TWI712315B - Method and system for recommending video - Google Patents
- ️Tue Dec 01 2020
TWI712315B - Method and system for recommending video - Google Patents
Method and system for recommending video Download PDFInfo
-
Publication number
- TWI712315B TWI712315B TW108138696A TW108138696A TWI712315B TW I712315 B TWI712315 B TW I712315B TW 108138696 A TW108138696 A TW 108138696A TW 108138696 A TW108138696 A TW 108138696A TW I712315 B TWI712315 B TW I712315B Authority
- TW
- Taiwan Prior art keywords
- user
- video
- audio
- preference
- similarity Prior art date
- 2019-10-25
Links
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The disclosure provides a method and system for recommending videos. The method includes: finding a plurality of first user groups from a plurality of existing users based on a preference, a historical viewing record, and a personal identity of a first user, wherein each first user group includes a plurality of first candidate users; retrieving a plurality of first videos that each first candidate user has watched; estimating a first similarity between each first video of each first candidate user and the first user, and generating a first video recommendation list; and in response to determining that a new video viewed by one of the first candidate users does not belong to the first video recommendation list, updating the first video recommendation list to be a second video recommendation list according to the new video, and provides the first user through a front end device.
Description
本發明是有關於一種多媒體檔案派送機制，且特別是有關於一種影音推荐方法及系統。 The present invention relates to a multimedia file delivery mechanism, and particularly relates to an audio-visual recommendation method and system.
隨著隨選視訊服務(Video on Demand，VoD)不斷地擴展，片庫系統內的影片持續大量增加。因此，對於本領域技術人員而言，如何藉由優化派片策略來將用戶可能最感興趣的影音推荐予用戶觀賞實為一項重要的議題。 With the continuous expansion of Video on Demand (VoD) services, the number of videos in the library system continues to increase substantially. Therefore, for those skilled in the art, it is an important issue how to recommend the videos that users may be most interested in by optimizing the distribution strategy.
有鑑於此，本發明提供一種影音推荐方法及系統，其可用於優化派片系統的派片策略，以向用戶推荐最適合的影音。 In view of this, the present invention provides a video and audio recommendation method and system, which can be used to optimize the distribution strategy of the film distribution system to recommend the most suitable video and audio to the user.
本發明提供一種影音推荐方法，包括：基於一第一用戶的一偏好、一歷史觀看記錄及一個人身分從多個既有用戶中找出多個第一用戶分群，其中各第一用戶分群包括多個第一候選用戶；取得各第一候選用戶所觀看過的多個第一影音；估計各第一候選 用戶的各第一影音與第一用戶的一第一相似度，並據以產生一第一影音推荐清單；反應於判定前述第一候選用戶的其中之一所觀看的一新影音不屬於第一影音推荐清單，依據新影音將第一影音推荐清單更新為一第二影音推荐清單，並透過一前端設備提供予第一用戶。 The present invention provides an audio-visual recommendation method, including: finding a plurality of first user groups from a plurality of existing users based on a preference of a first user, a historical viewing record and a person's identity, wherein each first user group includes multiple First candidate users; obtain multiple first videos watched by each first candidate user; estimate each first candidate A first similarity between each first video of the user and the first user, and a first video recommendation list is generated based on it; in response to determining that a new video watched by one of the first candidate users does not belong to the first The audio-visual recommendation list is updated to a second audio-visual recommendation list according to the new audio-visual recommendation list, and provided to the first user through a front-end device.
本發明提供一種影音推荐系統，包括前端設備及後端設備。前端設備屬於一第一用戶。後端設備經配置以：基於第一用戶的一偏好、一歷史觀看記錄及一個人身分從多個既有用戶中找出多個第一用戶分群，其中各第一用戶分群包括多個第一候選用戶；取得各第一候選用戶所觀看過的多個第一影音；估計各第一候選用戶的各第一影音與第一用戶的一第一相似度，並據以產生一第一影音推荐清單；反應於判定前述第一候選用戶的其中之一所觀看的一新影音不屬於第一影音推荐清單，依據新影音將第一影音推荐清單更新為一第二影音推荐清單，並透過前端設備提供予第一用戶。 The invention provides an audio-visual recommendation system, which includes front-end equipment and back-end equipment. The front-end equipment belongs to a first user. The back-end device is configured to: find a plurality of first user groups from a plurality of existing users based on a preference of the first user, a historical viewing record and a person's identity, wherein each first user group includes a plurality of first candidates Users; obtain a plurality of first video and audio watched by each first candidate user; estimate a first similarity between each first video and audio of each first candidate user and the first user, and generate a first video and audio recommendation list accordingly ; In response to determining that a new video watched by one of the first candidate users does not belong to the first video recommendation list, the first video recommendation list is updated to a second video recommendation list based on the new video and audio, and it is provided through the front-end device To the first user.
基於上述，本發明的方法及系統可令呈現予第一用戶的第一/第二影音推荐清單更為適合第一用戶，從而讓第一用戶可較輕易地從第一/第二影音推荐清單中找到有興趣的影音進行觀賞。 Based on the above, the method and system of the present invention can make the first/second audiovisual recommendation list presented to the first user more suitable for the first user, so that the first user can more easily select the first/second audiovisual recommendation list Find interesting videos to watch.
為讓本發明的上述特徵和優點能更明顯易懂，下文特舉實施例，並配合所附圖式作詳細說明如下。 In order to make the above-mentioned features and advantages of the present invention more comprehensible, the following specific embodiments are described in detail in conjunction with the accompanying drawings.
100:影音推荐系統 100: Audio-visual recommendation system
110:前端設備 110: front-end equipment
120:後端設備 120: back-end equipment
S210~S240:步驟 S210~S240: steps
圖1是依據本發明之一實施例繪示的影音推荐系統示意圖。 FIG. 1 is a schematic diagram of an audio-visual recommendation system according to an embodiment of the present invention.
圖2是依據本發明之一實施例繪示的影音推荐方法流程圖。 Fig. 2 is a flowchart of an audio-visual recommendation method according to an embodiment of the present invention.
請參照圖1，其是依據本發明之一實施例繪示的影音推荐系統示意圖。如圖1所示，影音推荐系統100包括前端設備110及後端設備120。在不同的實施例中，前端設備110例如是佈建於一般用戶家庭或其他類似場所內的VOD裝置，其可用以控制相關電視、投影機等裝置向用戶呈現用戶所選的影音介面或內容，但可不限於此。此外，後端設備120可連接於前端設備110，並可以是用於提供VOD影音服務的影音平台伺服器。一般而言，前端設備110的用戶可透過遙控器或其他控制裝置在前端設備110所提供的使用者介面中選擇所欲觀看的影音(例如電影、節目等)，而後端設備120即可相應地將用戶所選的影音內容發送至前端設備110，以由前端設備110透過電視、投影機等裝置將上述影音內容呈現予用戶觀賞。 Please refer to FIG. 1, which is a schematic diagram of an audio-visual recommendation system according to an embodiment of the present invention. As shown in FIG. 1, the audio-visual recommendation system 100 includes a front-end device 110 and a back-end device 120. In different embodiments, the front-end equipment 110 is, for example, a VOD device deployed in a general user’s home or other similar places. It can be used to control related TVs, projectors and other devices to present the user’s selected audio-visual interface or content to the user. But it is not limited to this. In addition, the back-end device 120 may be connected to the front-end device 110, and may be an audio-visual platform server for providing VOD audio-visual services. Generally speaking, the user of the front-end equipment 110 can select the desired video and audio (such as movies, programs, etc.) in the user interface provided by the front-end equipment 110 through the remote control or other control devices, and the back-end equipment 120 can respond accordingly. The audio-visual content selected by the user is sent to the front-end equipment 110, so that the front-end equipment 110 presents the above-mentioned audio-visual content to the user for viewing through a TV, a projector, and other devices.
在本發明的實施例中，後端設備120可用於執行本發明提出的影音推荐方法，以下將作詳細說明。 In the embodiment of the present invention, the back-end device 120 can be used to execute the audio-visual recommendation method proposed by the present invention, which will be described in detail below.
請參照圖2，其是依據本發明之一實施例繪示的影音推荐方法流程圖。本實施例的方法可由圖1的後端設備120執行，以下即搭配圖1所示的元件說明圖2各步驟的細節。 Please refer to FIG. 2, which is a flowchart of an audio-visual recommendation method according to an embodiment of the present invention. The method of this embodiment can be executed by the back-end device 120 in FIG. 1. The details of each step in FIG. 2 will be described below with the components shown in FIG. 1.
首先，在步驟S210中，後端設備120可基於第一用戶的偏好、歷史觀看記錄及個人身分從多個既有用戶中找出多個第一用戶分群，其中各第一用戶分群包括多個第一候選用戶。在本發明的實施例中，上述第一用戶例如是前端設備110所對應的用戶，而既有用戶則例如是使用後端設備120所提供服務的所有用戶，但可不限於此。 First, in step S210, the back-end device 120 may find multiple first user groups from multiple existing users based on the first user’s preferences, historical viewing records, and personal identity, where each first user group includes multiple The first candidate user. In the embodiment of the present invention, the above-mentioned first user is, for example, the user corresponding to the front-end device 110, and the existing users are, for example, all users who use the service provided by the back-end device 120, but it is not limited to this.
在第一實施例中，上述第一用戶分群中對應於第一用戶偏好的一者可略稱為偏好分群，而後端設備120找出屬於偏好分群的多個第一候選用戶的手段可例示如下。 In the first embodiment, one of the above-mentioned first user groupings corresponding to the first user preference may be referred to as a preference grouping, and the back-end device 120 finds out a plurality of first candidate users belonging to the preference grouping may be exemplified as follows .
在第一實施例中，後端設備120可取得第一用戶的網路瀏覽記錄，並基於網路瀏覽記錄所包括的多個網頁取得多個偏好關鍵子。上述網路瀏覽記錄例如是第一用戶所使用的網頁瀏覽器的搜尋記錄、我的最愛、cookie、書籤或是可用爬蟲方式所取得的社群網路的相關登錄資訊等，但不限於此。之後，後端設備120可將各筆網路瀏覽記錄中與影音相關的關鍵子作為偏好關鍵子，例如「科技」、「美國」等，但不限於此。 In the first embodiment, the back-end device 120 may obtain the web browsing history of the first user, and obtain multiple preference keys based on multiple web pages included in the web browsing history. The aforementioned Internet browsing records are, for example, search records of the web browser used by the first user, favorites, cookies, bookmarks, or social network related login information obtained by crawling methods, but not limited to this. After that, the back-end device 120 may use the key elements related to the video and audio in each web browsing record as the preference keys, such as "Technology", "United States", etc., but not limited to this.
之後，後端設備120可取得各偏好關鍵子所對應的檔案數量及檔案時間，並據以估計各偏好關鍵子的一偏好分數。舉例而言，在一實施例中，假設某偏好關鍵子出現於某網頁內容中，則後端設備120可計算此偏好關鍵子在此網頁內容(或網頁檔案)中的出現次數，並將此出現次數除以網頁內容的總字數，以得到此偏好關鍵子的詞頻。另外，若所考慮的偏好關鍵子為英文，則 可先去除其中的介係詞再進行上述操作，但可不限於此。接著，後端設備120可針對上述網頁內容相關的時間(例如搜尋時間、檔案時間等)對此網頁內容分配一檔案時間權重值(其可由設計者依需求而定)。例如，若上述網頁內容的時間點在三個月內，即可例示地設定為1.0，但可不限於此。 After that, the back-end device 120 can obtain the number of files and the file time corresponding to each preference key, and estimate a preference score of each preference key accordingly. For example, in one embodiment, assuming that a certain preference key appears in a certain webpage content, the back-end device 120 may calculate the number of occurrences of this preference key in the webpage content (or webpage file), and calculate the Divide the number of occurrences by the total number of words in the web content to get the word frequency of this preference key. In addition, if the considered preference key is English, then You can remove the prepositions before performing the above operations, but it is not limited to this. Then, the back-end device 120 can assign a file time weight value (which can be determined by the designer according to requirements) to the time (such as search time, file time, etc.) related to the web page content. For example, if the time point of the above web content is within three months, it can be set to 1.0 as an example, but it is not limited to this.
基此，第一用戶對於某偏好關鍵子的偏好分數可概略表徵為「(AVG(詞頻/檔案數)+檔案時間權重值+協同推荐權重值」，其中AVG代表取平均的運算子，而協同推荐權重值的定義將在之後詳述，此處可暫時理解為0。基此，在一實施例中，對於第一用戶而言，偏好關鍵子「科技」的偏好分數例如是「0.6+1.0+0.0=1.6」。 Based on this, the first user’s preference score for a certain preference key can be roughly represented as "(AVG (word frequency/number of files) + file time weight value + collaborative recommendation weight value", where AVG represents the average operator, and the collaborative The definition of the recommended weight value will be described in detail later, which can be temporarily understood as 0. Based on this, in one embodiment, for the first user, the preference score of the preference key sub-"technology" is, for example, "0.6+1.0 +0.0=1.6".
之後，後端設備120可估計各既有用戶對於各偏好關鍵子的參考偏好分數，而其作法可參照先前實施例的教示。舉例而言，對於一既有用戶B，其對於偏好關鍵子「科技」的參考偏好分數例如是「0.33+1.0+0.0=1.33」。舉另一例而言，對於一既有用戶C，其對於偏好關鍵子「科技」的參考偏好分數例如是「0.4+1.0+0.0=1.4」。在一實施例中，第一用戶、既有用戶B及C對於「科技」、「美國」等偏好關鍵子的偏好分數/參考偏好分數可例示如下表1。 After that, the back-end device 120 can estimate the reference preference score of each existing user for each preference key, and the method can refer to the teaching of the previous embodiment. For example, for an existing user B, his reference preference score for the preference key sub-"technology" is, for example, "0.33+1.0+0.0=1.33". For another example, for an existing user C, the reference preference score for the preference key sub-"technology" is, for example, "0.4+1.0+0.0=1.4". In one embodiment, the preference scores/reference preference scores of the first user and the existing users B and C for preference keys such as "Technology" and "United States" can be illustrated in Table 1 below.
之後，後端設備120可基於第一用戶的各偏好關鍵子的偏好分數與各既有用戶對於各偏好關鍵子的參考偏好分數估計第一用戶與各既有用戶之間的第一偏好相似度。在第一實施例中，後端設備120例如可基於歐式距離、餘弦公式、k-means演算法等手段來計算上述第一偏好相似度。舉例而言，假設所採用的是餘弦公式，則第一用戶與既有用戶B之間的第一偏好相似度可計算為：「 
接著，後端設備120可基於各既有用戶對應的第一偏好相似度排序前述既有用戶，並將排序在前的預設數量個既有用戶歸類至偏好分群，以作為偏好分群中的前述第一候選用戶。為便於說明，本發明各實施例中所提及的預設數量皆以N代稱，但在不同的情境中，各實施例所考慮的預設數量可以不同。基此，此處技術手段可概略理解為將既有用戶中具有較高第一偏好相似度的N者歸類至偏好分群中，以作為偏好分群中的第一候選用戶。 Then, the back-end device 120 may sort the aforementioned existing users based on the first preference similarity corresponding to each existing user, and classify the preset number of existing users that are ranked first into the preference group as the preference group. The aforementioned first candidate user. For ease of description, the preset numbers mentioned in the embodiments of the present invention are all referred to as N, but in different situations, the preset numbers considered in the embodiments may be different. Based on this, the technical means here can be roughly understood as categorizing N users with higher first preference similarity among existing users into preference groups, as the first candidate users in the preference groups.
在第二實施例中，上述第一用戶分群中對應於第一用戶歷史觀看記錄的一者可略稱為觀看分群，而後端設備120找出屬於觀看分群的多個第一候選用戶的手段可例示如下。 In the second embodiment, one of the above-mentioned first user groupings corresponding to the first user's historical viewing records can be referred to as a viewing group, and the back-end device 120 can find out the multiple first candidate users belonging to the viewing grouping. Examples are as follows.
在第二實施例中，第一用戶的歷史觀看記錄可包括第一用戶觀看過的多個歷史影音(例如曾使用前端設備110向後端設備120存取過的電影)。在此情況下，後端設備120可取得前述歷史影音所屬的多個影音分類，並統計前述歷史影音在各影音分類中的一歷史影音數量。舉例而言，假設在第一用戶看過的歷史影音中，有7部屬於「冒險動作片」，而其中3部屬於「經典冒險動作片」的影音分類，有1部屬於「動作喜劇」的影音分類，有3部屬於「動作驚悚片」的影音分類。在此情況下，「經典冒險動作片」、「動作喜劇」及「動作驚悚片」等影音分類所對應的歷史影音數量可分別例如是3、1、3。 In the second embodiment, the historical viewing record of the first user may include a plurality of historical videos and audios viewed by the first user (for example, movies that have been accessed by the front-end device 110 to the back-end device 120). In this case, the back-end device 120 can obtain multiple video and audio categories to which the aforementioned historical videos belong, and count the number of historical videos in each of the aforementioned historical videos and audio categories. For example, suppose that among the historical videos and audios watched by the first user, 7 belong to "adventure action movies", 3 of which belong to the category of "classic action adventure movies", and 1 of them belong to "action comedy". Category, there are 3 video and audio categories belonging to "action thriller". In this case, the number of historical videos corresponding to the video categories such as "classic adventure action movie", "action comedy", and "action thriller" can be 3, 1, 3, respectively.
之後，後端設備120可取得各既有用戶所觀看過的多個參考影音，並累計各既有用戶的前述參考影音在各影音分類中的一參考影音數量。舉例而言，假設在既有用戶B看過的參考影音中，有10部屬於「冒險動作片」，而其中2部屬於「經典冒險動作片」的影音分類，有4部屬於「動作喜劇」的影音分類，有4部屬於「動作驚悚片」的影音分類。在此情況下，「經典冒險動作片」、「動作喜劇」及「動作驚悚片」等影音分類所對應的參考影音數量可分別例如是2、4、4。在第二實施例中，第一用戶、既有用戶B及C對於「冒險動作片」、「動作喜劇」及「動作驚悚片」等影音分類的歷史影音數量/參考影音數量可例示如下表2。 After that, the back-end device 120 can obtain multiple reference videos watched by each existing user, and accumulate a number of reference videos in each video category of the aforementioned reference videos of each existing user. For example, suppose that among the reference videos that the existing user B has watched, 10 belong to "adventure action movies", 2 of which belong to the video category of "classic action adventure movies", and 4 belong to "action comedy" There are 4 video and audio categories that belong to "action thriller". In this case, the number of reference videos corresponding to the video categories such as "classic adventure action movie", "action comedy" and "action thriller" may be 2, 4, and 4, respectively. In the second embodiment, the first user, existing users B and C, for "adventure action movie", "action comedy" and "action thriller" and other audiovisual categories of the historical audiovisual quantity/reference audiovisual quantity can be illustrated in Table 2 below .
之後，後端設備120可基於前述歷史影音在各影音分類中的歷史影音數量及各既有用戶的前述參考影音在各影音分類中的參考影音數量估計第一用戶與各既有用戶之間的一第一分類相似度。在第二實施例中，後端設備120同樣可於歐式距離、餘弦公式、k-means演算法等手段來計算上述第一分類相似度。舉例而言，假設所採用的是餘弦公式，則第一用戶與既有用戶B之間的第一分類相似度可計算為：「第一分類相似度(第一用戶，既有用戶B)= 
接著，後端設備120可基於各既有用戶對應的第一分類相似度排序前述既有用戶，並將排序在前的預設數量個既有用戶歸類至觀看分群，以作為觀看分群中的前述第一候選用戶。基此，此處技術手段可概略理解為將既有用戶中具有較高第一分類相似度的N者歸類至觀看分群中，以作為觀看分群中的第一候選用戶。 Then, the back-end device 120 may sort the aforementioned existing users based on the first classification similarity corresponding to each existing user, and classify the preset number of existing users that are ranked first into the viewing group as the viewing group. The aforementioned first candidate user. Based on this, the technical means here can be roughly understood as categorizing N persons with higher first classification similarity among existing users into the viewing group as the first candidate user in the viewing group.
在第三實施例中，上述第一用戶分群中對應於第一用戶個人身分的一者可略稱為身分分群，而後端設備120找出屬於身 分分群的多個第一候選用戶的手段可例示如下。 In the third embodiment, one of the above-mentioned first user groupings corresponding to the personal identity of the first user may be referred to as an identity group, and the back-end device 120 finds out the identity group. The means for grouping a plurality of first candidate users can be exemplified as follows.
在第三實施例中，第一用戶的個人身分包括可多個第一用戶資料，包括但不限於身高、體重等。在此情況下，後端設備120可取得各既有用戶的多個參考用戶資料，其中前述參考用戶資料對應於前述第一用戶資料。亦即，後端設備120可取得各既有用戶的身體、體重等資料，但不限於此。在第三實施例中，第一用戶的第一用戶資料，以及既有用戶B、既有用戶C的參考用戶資料可例示如下表3。 In the third embodiment, the personal identity of the first user includes multiple first user data, including but not limited to height, weight, and so on. In this case, the back-end device 120 can obtain multiple reference user data of each existing user, where the aforementioned reference user data corresponds to the aforementioned first user data. That is, the back-end device 120 can obtain the body and weight of each existing user, but it is not limited to this. In the third embodiment, the first user profile of the first user and the reference user profile of the existing user B and the existing user C can be illustrated in Table 3 below.
之後，後端設備120可基於第一用戶的前述第一用戶資料及各既有用戶的前述參考用戶資料估計第一用戶與各既有用戶之間的一第一身分相似度。在第三實施例中，後端設備120計算第一用戶與各既有用戶之間的第一身分相似度的方式可參考第一、第二實施例中所教示的手段，於此不另贅述。 After that, the back-end device 120 may estimate a first identity similarity between the first user and each existing user based on the aforementioned first user profile of the first user and the aforementioned reference user profile of each existing user. In the third embodiment, the method of calculating the first identity similarity between the first user and each existing user by the back-end device 120 can refer to the methods taught in the first and second embodiments, which will not be repeated here. .
接著，後端設備120可基於各既有用戶對應的第一身分相似度排序前述既有用戶，並將排序在前的預設數量個既有用戶歸類至身分分群，以作為身分分群中的前述第一候選用戶。基此， 此處技術手段可概略理解為將既有用戶中具有較高第一身分相似度的N者歸類至身分分群中，以作為身分分群中的第一候選用戶。 Then, the back-end device 120 may sort the aforementioned existing users based on the first identity similarity corresponding to each existing user, and classify the preset number of existing users ranked first in the identity group as the identity group. The aforementioned first candidate user. Based on this, The technical means here can be roughly understood as categorizing N persons with high first identity similarity among existing users into an identity group as the first candidate user in the identity group.
由第一、第二及第三實施例可知，偏好分群、觀看分群及身分分群可個別包括N個第一候選用戶(亦，共有3×N個第一候選用戶)，而所述3×N個第一候選用戶可概略理解為在既有用戶中與所考慮的第一用戶最為相似的一部分用戶，但其僅用以舉例，並非用以限定本發明可能的實施方式。 It can be seen from the first, second and third embodiments that the preference grouping, viewing grouping, and identity grouping can each include N first candidate users (also, there are 3×N first candidate users in total), and the 3×N A first candidate user can be roughly understood as a part of the existing users most similar to the considered first user, but it is only used as an example and is not used to limit the possible implementation of the present invention.
在取得偏好分群、觀看分群及身分分群個別的N個第一候選用戶之後，在步驟S220中，後端設備120可取得各第一候選用戶所觀看過的多個第一影音。亦即，後端設備120可取得與第一用戶最相似的一部分用戶所觀看過的所有影音。 After obtaining the N first candidate users of the preference group, the viewing group and the identity group, in step S220, the back-end device 120 may obtain a plurality of first video and audio watched by each first candidate user. That is, the back-end device 120 can obtain all the videos watched by a part of users who are most similar to the first user.
之後，在步驟S230中，後端設備120可估計各第一候選用戶的各第一影音與第一用戶的第一相似度，並據以產生第一影音推荐清單。 After that, in step S230, the back-end device 120 may estimate the first similarity between each first video and audio of each first candidate user and the first user, and generate a first video and audio recommendation list accordingly.
在第一實施例中，後端設備120可取得各第一影音對應於各偏好關鍵子(例如「科技」、「美國」)的一第一影音偏好分數。為便於說明，以下將以「夏洛克」及「犯罪現場調查」作為既有用戶C曾觀看過的第一影音的例子，而其個別的第一影音偏好分數可例示如下表4。 In the first embodiment, the back-end device 120 can obtain a first audio-visual preference score of each first audio-visual corresponding to each preference key (for example, "Technology", "United States"). For the convenience of description, the following will take "Sherlock" and "Crime Scene Investigation" as examples of the first video and audio that the existing user C has watched, and their individual first video and audio preference scores can be illustrated in Table 4 below.
此外，依據表1的內容可知，第一用戶對應「科技」、「美國」等偏好關鍵子的偏好分數分別是1.6、1。在此情況下，後端設備120可基於各第一影音對應於各偏好關鍵子的第一影音偏好分數及第一用戶對應各偏好關鍵子的偏好分數估計第一用戶與各第一影音之間的第一相似度。在第一實施例中，假設所採用的是餘弦公式，則第一用戶與第一影音「夏洛克」之間的第一相似度可表徵為：「 
之後，後端設備120可依據各第一影音對應的第一相似度排序前述第一影音，並基於前述第一影音中排序在前的預設數量個第一影音產生第一影音推荐清單。亦即，後端設備120可將第一實施例中具較高第一相似度的N個第一影音歸類至第一影音推荐清單中。 After that, the back-end device 120 may sort the first video and audio according to the first similarity corresponding to each first video and audio, and generate a first video and audio recommendation list based on a preset number of first audio and video that are ranked first among the first videos. That is, the back-end device 120 can classify the N first videos with a higher first similarity in the first embodiment into the first video recommendation list.
此外，在第二實施例中，後端設備120可統計前述歷史影音的多個特定關鍵子個別的一第一關鍵子次數。為便於說明，以下假設所考慮的特定關鍵子為「推理」及「喜劇」。在此情況下，假設第一用戶所觀賞過的歷史影音中共有4個對應於特定關鍵子 「推理」的影音，並共有3個對應於特定關鍵子「喜劇」的影音。在此情況下，對於第一用戶而言，前述特定關鍵子個別的第一關鍵子次數可例示如下表5。 In addition, in the second embodiment, the back-end device 120 can count the number of individual first key sub-times of the multiple specific key sub-units of the aforementioned historical video and audio. For the convenience of explanation, the specific keys considered in the following hypotheses are "reasoning" and "comedy". In this case, suppose that there are 4 historical videos watched by the first user corresponding to specific key sub There are 3 videos corresponding to the specific key sub-"Comedy". In this case, for the first user, the individual first key times of the aforementioned specific keys can be illustrated in Table 5 below.
之後，後端設備120可取得各第一影音(例如「夏洛克」、「犯罪現場調查」)對應於各特定關鍵子(例如，「推理」及「喜劇」)的第二關鍵子次數。在第二實施例中，各第一影音對應於各特定關鍵子的第二關鍵子次數可例示如下表6。 After that, the back-end device 120 can obtain the second key number of each first video (for example, "Sherlock", "crime scene investigation") corresponding to each specific key (for example, "inference" and "comedy"). In the second embodiment, the second key times of each first video and audio corresponding to each specific key can be illustrated in Table 6 below.
接著，後端設備120可基於前述歷史影音的前述特定關鍵子個別的第一關鍵子次數及各第一影音對應於各特定關鍵子的第二關鍵子次數估計第一用戶與各第一影音的第一相似度。。在第二實施例中，假設所採用的是餘弦公式，則第一用戶與第一影音「夏洛克」之間的第一相似度可表徵為：「 
之後，後端設備120可依據各第一影音對應的第一相似度排序前述第一影音，並基於前述第一影音中排序在前的預設數量個第一影音產生第一影音推荐清單。亦即，後端設備120可將第二實施例中具較高第一相似度的N個第一影音歸類至第一影音推荐清單中。 After that, the back-end device 120 may sort the first video and audio according to the first similarity corresponding to each first video and audio, and generate a first video and audio recommendation list based on a preset number of first audio and video that are ranked first among the first videos. That is, the back-end device 120 can classify the N first videos with higher first similarity in the second embodiment into the first video recommendation list.
此外，在第三實施例中，後端設備120可將身分分群中的第一候選用戶中與第一用戶最為相似的一或多者所觀看過的N個第一影音歸類至第一影音推荐清單中。 In addition, in the third embodiment, the back-end device 120 may classify the N first videos viewed by one or more of the first candidate users in the identity group that are most similar to the first user into the first videos. Recommended list.
由上可概略看出，第一影音推荐清單中可包括由第一、第二、第三實施例所個別得出的N個第一影音，故第一影音推荐清單中共可包括3×N個第一影音。在一實施例中，後端設備120還可進一步從所述3×N個第一影音取出與第一用戶最為相關的N者，並將其餘的第一影音從第一影音推荐清單移除，但可不限於此。 It can be roughly seen from the above that the first audio-visual recommendation list can include N first audio-visuals individually derived from the first, second, and third embodiments, so the first audio-visual recommendation list can include 3×N in total The first video. In an embodiment, the back-end device 120 may further retrieve the N most relevant to the first user from the 3×N first videos, and remove the remaining first videos from the first video recommendation list. But it is not limited to this.
在依據以上教示產生第一影音推荐清單之後，在步驟S240中，反應於判定前述第一候選用戶的其中之一所觀看的新影音不屬於第一影音推荐清單，後端設備120可依據新影音將第一影音推荐清單更新為第二影音推荐清單，並透過前端設備110提供予第一用戶。 After the first audiovisual recommendation list is generated according to the above teachings, in step S240, in response to determining that the new audiovisual watched by one of the first candidate users does not belong to the first audiovisual recommendation list, the back-end device 120 may follow the new audiovisual recommendation list. The first audio-visual recommendation list is updated to the second audio-visual recommendation list and provided to the first user through the front-end device 110.
在一實施例中，後端設備120可取得新影音的多個新關 鍵子。之後，反應於判定前述新關鍵子的第一新關鍵子對應於前述偏好關鍵的第一偏好關鍵子，修正第一偏好關鍵子的偏好分數。 In one embodiment, the back-end device 120 can obtain multiple new levels of new audio and video. Keys. Then, in response to determining that the first new key element of the aforementioned new key element corresponds to the first preference key element of the aforementioned preference key, the preference score of the first preference key element is corrected.
為便於說明，以下假設所述新影音為「超人」，而其所對應的新關鍵子例如包括「科技」，但可不限於此。在此情況下，由於新關鍵子「科技」對應於表1中的偏好關鍵子「科技」，故後端設備120可相應地修正偏好關鍵子的偏好分數。 For ease of description, the following assumes that the new video is "Superman", and the corresponding new key includes, for example, "Technology", but it is not limited to this. In this case, since the new key sub "technology" corresponds to the preference key sub "technology" in Table 1, the back-end device 120 can modify the preference score of the preference key accordingly.
具體而言，由第一實施例中的教示可知，偏好關鍵子「科技」的偏好分數原本為1.6，而其是基於「(AVG(詞頻/檔案數)+檔案時間權重值+協同推荐權重值」而得出。基此，當後端設備120欲修正偏好關鍵子「科技」的偏好分數時，可藉由調整上式中的「協同推荐權重值」而為之。舉例而言，後端設備120可將偏好關鍵子「科技」的「協同推荐權重值」由0調整為0.1。在此情況下，偏好關鍵子「科技」的偏好分數將從1.6被修正為1.7，但本發明可不限於此。 Specifically, according to the teaching in the first embodiment, the preference score of the preference key sub "Technology" was originally 1.6, and it was based on "(AVG (word frequency/number of files) + file time weight value + collaborative recommendation weight value Based on this, when the back-end device 120 wants to modify the preference score of the preference key sub-"technology", it can be done by adjusting the "collaborative recommendation weight value" in the above formula. For example, the back-end The device 120 may adjust the "collaborative recommendation weight value" of the preference key sub-"technology" from 0 to 0.1. In this case, the preference score of the preference key sub-"technology" will be revised from 1.6 to 1.7, but the present invention is not limited to this.
此外，反應於判定新影音屬於前述影音分類中的第一影音分類，後端設備120還可累加第一影音分類的歷史影音數量。承上例，在新影音為「超人」的情況下，其可經判定為屬於表2中的「經典冒險動作片」。因此，後端設備120可將「經典冒險動作片」的歷史影音數量由3累加為4，但本發明可不限於此。 In addition, in response to determining that the new video and audio belong to the first video and audio category in the aforementioned video and audio classification, the back-end device 120 may also accumulate the number of historical videos of the first video and audio category. Continuing the above example, if the new movie is "Superman", it can be judged to belong to the "Classic Adventure Action Movie" in Table 2. Therefore, the back-end device 120 can accumulate the number of historical videos of the "classic adventure action movie" from 3 to 4, but the invention is not limited to this.
之後，後端設備120可依據第一偏好關鍵子(例如「科技」)修正後的偏好分數(例如1.7)及第一影音分類(例如「經 典冒險動作片」)累加後的歷史影音數量(例如4)重新產生偏好分群、觀看分群及身分分群，其中偏好分群、觀看分群及身分分群個別包括多個第二候選用戶。 After that, the back-end device 120 may modify the preference score (for example, 1.7) and the first audiovisual category (for example, "Classic Adventure Action Movie") accumulated historical video and audio quantity (for example, 4) to regenerate preference group, viewing group and identity group, where preference group, viewing group and identity group each include multiple second candidate users.
概略而言，後端設備120可基於第一、第二及第三實施例先前的教示內容而再次從既有用戶中取得偏好分群、觀看分群及身分分群，惟所考慮的第一偏好關鍵子(例如「科技」)的偏好分數係修正後的分數，且第一影音分類(例如「經典冒險動作片」)的歷史影音數量是累加後的數值。 Roughly speaking, the back-end device 120 can again obtain preference grouping, viewing grouping and identity grouping from existing users based on the previous teaching content of the first, second, and third embodiments, but the first preference key is considered The preference score of (for example, "Technology") is a modified score, and the number of historical videos in the first video category (for example, "Classic Action Adventure") is the accumulated value.
具體來說，後端設備120可基於第一用戶的各偏好關鍵子的(修正後)偏好分數與各既有用戶對於各偏好關鍵子的參考偏好分數估計第一用戶與各既有用戶之間的第二偏好相似度。第二偏好相似度可參照先前計算第一偏好相似度的方式，於此不另贅述。 Specifically, the back-end device 120 may estimate the relationship between the first user and each existing user based on the (modified) preference score of each preference key of the first user and the reference preference score of each existing user for each preference key. The second preference similarity. The second preference similarity can refer to the previous method of calculating the first preference similarity, which will not be repeated here.
之後，後端設備120可基於各既有用戶對應的第二偏好相似度排序前述既有用戶，並將排序在前的預設數量個既有用戶歸類至偏好分群，以作為偏好分群中的前述第二候選用戶。基此，此處技術手段可概略理解為將既有用戶中具有較高第二偏好相似度的N者歸類至偏好分群中，以作為(更新後)偏好分群中的第二候選用戶。 After that, the back-end device 120 may sort the aforementioned existing users based on the second preference similarity corresponding to each existing user, and classify the preset number of existing users that are ranked first into the preference group as the preference group. The aforementioned second candidate user. Based on this, the technical means here can be roughly understood as categorizing N users with higher second preference similarity among existing users into preference groups as the second candidate users in the (updated) preference group.
另外，後端設備120可基於前述歷史影音在各影音分類中的(累加後)歷史影音數量及各既有用戶的前述參考影音在各影音分類中的參考影音數量估計第一用戶與各既有用戶之間的一 第二分類相似度。第二分類相似度可參照先前計算第一分類相似度的方式，於此不另贅述。 In addition, the back-end device 120 may estimate the first user and each existing user based on the (accumulated) number of historical videos in each video category of the aforementioned historical videos and the number of reference videos of each existing user in each video category. One among users The second category similarity. The second category similarity can refer to the previous method of calculating the first category similarity, which will not be repeated here.
之後，後端設備120可基於各既有用戶對應的第二分類相似度排序前述既有用戶，並將排序在前的預設數量個既有用戶歸類至觀看分群，以作為觀看分群中的前述第二候選用戶。基此，此處技術手段可概略理解為將既有用戶中具有較高第二分類相似度的N者歸類至觀看分群中，以作為(更新後)觀看分群中的第二候選用戶。 After that, the back-end device 120 may sort the aforementioned existing users based on the second classification similarity corresponding to each existing user, and classify the preset number of existing users that are ranked first into the viewing group as the viewing group. The aforementioned second candidate user. Based on this, the technical means here can be roughly understood as categorizing N among the existing users with higher second classification similarity into the viewing group as the second candidate user in the (updated) viewing group.
在重新產生偏好分群、觀看分群及身分分群(其個別可包括N個第二候選用戶)之後，後端設備120可取得各第二候選用戶所觀看過的多個第二影音。接著，後端設備120可估計各第二候選用戶的各第二影音與第一用戶的一第二相似度，並據以產生第二影音推荐清單。 After regenerating the preference group, the viewing group, and the identity group (each of which may include N second candidate users), the backend device 120 may obtain a plurality of second videos viewed by each second candidate user. Then, the back-end device 120 can estimate a second similarity between each second video and audio of each second candidate user and the first user, and generate a second video and audio recommendation list accordingly.
概略而言，後端設備120可採用先前估計各第一候選用戶的各第一影音與第一用戶的第一相似度的方式來估計各第二候選用戶的各第二影音與第一用戶的第二相似度。 Roughly speaking, the back-end device 120 may estimate the first similarity between each first video and audio of each first candidate user and the first user to estimate each second video and audio of each second candidate user and the first user. The second degree of similarity.
具體來說，在第一實施例中，後端設備120可取得各第二影音對應於各偏好關鍵子的第二影音偏好分數，並基於各第二影音對應於各偏好關鍵子的第二影音偏好分數及第一用戶對應各偏好關鍵子的偏好分數估計第一用戶與各第二影音之間的第二相似度。相關細節可參照先前第一實施例中取得第一相似度的說明，於此不另贅述。 Specifically, in the first embodiment, the back-end device 120 can obtain the second audio-visual preference score of each second audio-visual corresponding to each preference key, and based on the second audio-visual preference score of each second audio-visual corresponding to each preference key. The preference score and the preference score of the first user corresponding to each preference key estimate the second similarity between the first user and each second video and audio. For related details, please refer to the description of obtaining the first similarity in the previous first embodiment, which will not be repeated here.
之後，後端設備120可依據各第二影音對應的第二相似度排序前述第二影音，並基於前述第二影音中排序在前的預設數量個第二影音產生第二影音推荐清單。亦即，後端設備120可將第一實施例中具較高第二相似度的N個第二影音歸類至第二影音推荐清單中。 After that, the back-end device 120 may sort the second video and audio according to the second similarity corresponding to each second video and audio, and generate a second video and audio recommendation list based on the preset number of second audio and video that are ranked first among the second videos. That is, the back-end device 120 can classify the N second videos with higher second similarity in the first embodiment into the second video recommendation list.
另外，在第二實施例中，後端設備120可取得各第二影音對應於各特定關鍵子的一第三關鍵子次數，並基於前述歷史影音的前述特定關鍵子個別的第一關鍵子次數及各第二影音對應於各特定關鍵子的第三關鍵子次數估計第一用戶與各第二影音的第二相似度。相關細節可參照先前第二實施例中取得第一相似度的說明，於此不另贅述。 In addition, in the second embodiment, the back-end device 120 may obtain a third key number corresponding to each specific key of each second video and audio, and based on the individual first key number of the specific key of the aforementioned historical video and audio. And the third key times of each second video and audio corresponding to each specific key to estimate the second similarity between the first user and each second video and audio. For related details, please refer to the description of obtaining the first similarity in the second embodiment, which will not be repeated here.
之後，後端設備120可依據各第二影音對應的第二相似度排序前述第二影音，並基於前述第二影音中排序在前的預設數量個第二影音產生第二影音推荐清單。亦即，後端設備120可將第二實施例中具較高第二相似度的N個第二影音歸類至第二影音推荐清單中。 After that, the back-end device 120 may sort the second video and audio according to the second similarity corresponding to each second video and audio, and generate a second video and audio recommendation list based on the preset number of second audio and video that are ranked first among the second videos. That is, the back-end device 120 can classify the N second videos with higher second similarity in the second embodiment into the second video recommendation list.
此外，在第三實施例中，後端設備120可將(重新產生的)身分分群中的第二候選用戶中與第一用戶最為相似的一或多者所觀看過的N個第二影音歸類至第二影音推荐清單中。 In addition, in the third embodiment, the back-end device 120 may return the N second videos viewed by one or more of the second candidate users in the (regenerated) identity group that are most similar to the first user. Class to the second audiovisual recommendation list.
由上可概略看出，第二影音推荐清單中可包括由第一、第二、第三實施例所個別得出的N個第二影音，故第二影音推荐清單中共可包括3×N個第二影音。在一實施例中，後端設備120 還可進一步從所述3×N個第二影音取出與第一用戶最為相關的N者，並將其餘的第二影音從第二影音推荐清單移除，但可不限於此。 It can be roughly seen from the above that the second audio-visual recommendation list can include N second audio-visuals individually derived from the first, second, and third embodiments, so the second audio-visual recommendation list can include a total of 3×N The second video. In one embodiment, the back-end device 120 It is also possible to further extract the N most relevant to the first user from the 3×N second videos, and remove the remaining second videos from the second video recommendation list, but it is not limited to this.
在依據上述教示取得第二影音推荐清單之後，後端設備120可將第二影音推荐清單提供予前端設備110，以透過前端設備110將第二影音推荐清單呈現予第一用戶參考。藉此，可令第一用戶查看與自身偏好/習慣/身分較為適配的第二影音推荐清單，從而可較輕易地從中找到有興趣的影音進行觀賞。從另一觀點而言，本發明還可理解為避免將第一用戶較不感興趣的影音推荐予第一用戶，從而可改善第一用戶的用戶體驗。 After obtaining the second audio-visual recommendation list according to the above-mentioned teaching, the back-end device 120 may provide the second audio-visual recommendation list to the front-end device 110 to present the second audio-visual recommendation list to the first user for reference through the front-end device 110. In this way, the first user can view the second audio-visual recommendation list that is more suitable for his own preferences/habits/identities, so that the interested audio-visual list can be found more easily to watch. From another point of view, the present invention can also be understood as avoiding recommending to the first user the video and audio that the first user is less interested in, thereby improving the user experience of the first user.
在一實施例中，若所考慮的第一用戶為影音推荐系統100的新用戶(亦即未有相關的歷史觀看記錄)，則後端設備120可僅基於第一、第三實施例的教示來產生第一、第二影音推荐清單，但本發明可不限於此。 In an embodiment, if the first user under consideration is a new user of the audio-visual recommendation system 100 (that is, there is no related historical viewing record), the back-end device 120 may only be based on the teachings of the first and third embodiments To generate the first and second audio-visual recommendation lists, but the present invention is not limited to this.
綜上所述，本發明可概略理解為包括兩個階段。在第一個階段中，本發明的方法及系統可先依據第一用戶的偏好、歷史觀看記錄及個人身分從既有用戶中找出與第一用戶較為相似的偏好分群、觀看分群及身分分群，再依據這些分群中各第一候選用戶所觀看過的第一影音產生第一影音推荐清單。之後，若判定某第一候選用戶所觀看的新影音不屬於第一影音推荐清單，則第二個階段可相應地被觸發。 In summary, the present invention can be roughly understood as including two stages. In the first stage, the method and system of the present invention can first find out preference groups, viewing groups and identity groups that are similar to the first user from existing users based on the first user’s preferences, historical viewing records and personal identity. , And then generate a first video and audio recommendation list based on the first video and audio watched by each first candidate user in these groups. Afterwards, if it is determined that the new video and audio watched by a certain first candidate user does not belong to the first video and audio recommendation list, the second stage can be triggered accordingly.
在第二個階段中，本發明的方法可依據新影音修正對應 的偏好關鍵子的偏好分數及/或累加對應的影音分類的歷史影音數量，並據以產生新的偏好分群、觀看分群及身分分群。之後，可再依據這些分群中各第二候選用戶所觀看過的第二影音產生第二影音推荐清單。 In the second stage, the method of the present invention can be corrected according to the new video and audio The preference score of the preference key and/or the number of historical videos of the corresponding video category are accumulated, and new preference groups, viewing groups and identity groups are generated accordingly. After that, a second video and audio recommendation list can be generated according to the second video and audio watched by each second candidate user in these groups.
如此一來，可令呈現予第一用戶的第一/第二影音推荐清單更為適合第一用戶。因此，當本發明應用於後端設備所提供的影音平台時，針對動輒數十萬部以上的影音，除了能夠進行有效過濾之外，還可自動學習用戶有興趣的影音，從而進行推荐。並且，本發明還可解決新用戶一開始沒有歷史觀看記錄時的推荐問題，並結合既有用戶，提供一個整合式的影音推荐清單。 In this way, the first/second audiovisual recommendation list presented to the first user can be more suitable for the first user. Therefore, when the present invention is applied to an audio-visual platform provided by a back-end device, in addition to effective filtering of audio and video with hundreds of thousands or more, it can also automatically learn the audio and video that users are interested in for recommendation. Moreover, the present invention can also solve the recommendation problem when a new user has no historical viewing record at the beginning, and combine with existing users to provide an integrated audio-visual recommendation list.
雖然本發明已以實施例揭露如上，然其並非用以限定本發明，任何所屬技術領域中具有通常知識者，在不脫離本發明的精神和範圍內，當可作些許的更動與潤飾，故本發明的保護範圍當視後附的申請專利範圍所界定者為準。 Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention. Anyone with ordinary knowledge in the technical field can make some changes and modifications without departing from the spirit and scope of the present invention. The scope of protection of the present invention shall be determined by the scope of the attached patent application.
S210~S240:步驟 S210~S240: steps
Claims (11)
一種影音推荐方法，包括： 基於一第一用戶的一偏好、一歷史觀看記錄及一個人身分從多個既有用戶中找出多個第一用戶分群，其中各該第一用戶分群包括多個第一候選用戶； 取得各該第一候選用戶所觀看過的多個第一影音； 估計各該第一候選用戶的各該第一影音與該第一用戶的一第一相似度，並據以產生一第一影音推荐清單； 反應於判定該些第一候選用戶的其中之一所觀看的一新影音不屬於該第一影音推荐清單，依據該新影音將該第一影音推荐清單更新為一第二影音推荐清單，並透過一前端設備提供予該第一用戶。 An audio-visual recommendation method, including: Finding a plurality of first user groups from a plurality of existing users based on a preference of a first user, a historical viewing record and a person's identity, wherein each of the first user groups includes a plurality of first candidate users; Obtaining a plurality of first videos and audios watched by each of the first candidate users; Estimating a first similarity between each of the first video and audio of each of the first candidate users and the first user, and generating a first video and audio recommendation list accordingly; In response to determining that a new video and audio watched by one of the first candidate users does not belong to the first video and audio recommendation list, the first video and audio recommendation list is updated to a second video and audio recommendation list according to the new video and audio, and through A front-end device is provided to the first user. 如申請專利範圍第1項所述的方法，其中該些第一用戶分群包括對應於該第一用戶的該偏好的一偏好分群，且所述方法包括： 取得該第一用戶的一網路瀏覽記錄，並基於該網路瀏覽記錄所包括的多個網頁取得多個偏好關鍵子； 取得各該偏好關鍵子所對應的檔案數量及檔案時間，並據以估計各該偏好關鍵子的一偏好分數； 估計各該既有用戶對於各該偏好關鍵子的一參考偏好分數； 基於該第一用戶的各該偏好關鍵子的該偏好分數與各該既有用戶對於各該偏好關鍵子的該參考偏好分數估計該第一用戶與各該既有用戶之間的一第一偏好相似度； 基於各該既有用戶對應的該第一偏好相似度排序該些既有用戶，並將排序在前的預設數量個既有用戶歸類至該偏好分群，以作為該偏好分群中的該些第一候選用戶。 The method according to claim 1, wherein the first user groups include a preference group corresponding to the preference of the first user, and the method includes: Obtain a web browsing record of the first user, and obtain multiple preference keys based on multiple web pages included in the web browsing record; Obtain the number of files and file time corresponding to each preference key, and estimate a preference score of each preference key accordingly; Estimate a reference preference score of each existing user for each preference key; Estimating a first preference between the first user and each existing user based on the preference score of each preference key of the first user and the reference preference score of each existing user for each preference key Similarity The existing users are sorted based on the first preference similarity corresponding to each of the existing users, and the preset number of existing users ranked first are classified into the preference group as the preference groups. The first candidate user. 如申請專利範圍第2項所述的方法，其中估計各該第一候選用戶的各該第一影音與該第一用戶的該第一相似度，並據以產生該第一影音推荐清單的步驟包括： 取得各該第一影音對應於各該偏好關鍵子的一第一影音偏好分數； 基於各該第一影音對應於各該偏好關鍵子的該第一影音偏好分數及該第一用戶對應各該偏好關鍵子的該偏好分數估計該第一用戶與各該第一影音之間的該第一相似度； 依據各該第一影音對應的該第一相似度排序該些第一影音，並基於該些第一影音中排序在前的預設數量個第一影音產生該第一影音推荐清單。 The method according to item 2 of the scope of patent application, wherein the step of estimating the first similarity between each of the first video and audio of each of the first candidate users and the first user, and generating the first video and audio recommendation list accordingly include: Obtain a first audio-visual preference score corresponding to each of the preference keys for each of the first audio-visuals; Based on the first audio-visual preference score of each of the first audio-visual corresponding to each of the preference keys and the preference score of the first user corresponding to each of the preference keys, the estimation of the difference between the first user and each of the first audio-visual First similarity The first video and audio are sorted according to the first similarity corresponding to each of the first video and audio, and the first video and audio recommendation list is generated based on a preset number of first video and audio that are ranked first among the first videos. 如申請專利範圍第2項所述的方法，其中該第一用戶的該歷史觀看記錄包括該第一用戶觀看過的多個歷史影音，且該些第一用戶分群包括對應於該歷史觀看記錄的一觀看分群，且所述方法包括： 取得該些歷史影音所屬的多個影音分類，並統計該些歷史影音在各該影音分類中的一歷史影音數量； 取得各該既有用戶所觀看過的多個參考影音，並累計各該既有用戶的該些參考影音在各該影音分類中的一參考影音數量； 基於該些歷史影音在各該影音分類中的該歷史影音數量及各該既有用戶的該些參考影音在各該影音分類中的該參考影音數量估計該第一用戶與各該既有用戶之間的一第一分類相似度； 基於各該既有用戶對應的該第一分類相似度排序該些既有用戶，並將排序在前的預設數量個既有用戶歸類至該觀看分群，以作為該觀看分群中的該些第一候選用戶。 The method according to claim 2, wherein the historical viewing record of the first user includes a plurality of historical videos watched by the first user, and the first user groups include the historical viewing records corresponding to the historical viewing record. One viewing grouping, and the method includes: Obtain multiple video and audio categories to which the historical videos belong, and count the number of historical videos in each video and audio category; Acquire multiple reference videos watched by each existing user, and accumulate the number of reference videos of each existing user in each video category; Based on the number of historical videos in each video category of the historical videos and the number of reference videos of each existing user in each video category, it is estimated that the first user and each of the existing users A first category similarity between; The existing users are sorted based on the similarity of the first category corresponding to each of the existing users, and the preset number of existing users ranked first are classified into the viewing group as the viewing group. The first candidate user. 如申請專利範圍第4項所述的方法，其中估計各該第一候選用戶的各該第一影音與該第一用戶的該第一相似度，並據以產生該第一影音推荐清單的步驟包括： 統計該些歷史影音的多個特定關鍵子個別的一第一關鍵子次數； 取得各該第一影音對應於各該特定關鍵子的一第二關鍵子次數； 基於該些歷史影音的該些特定關鍵子個別的該第一關鍵子次數及各該第一影音對應於各該特定關鍵子的該第二關鍵子次數估計該第一用戶與各該第一影音的該第一相似度； 依據各該第一影音對應的該第一相似度排序該些第一影音，並基於該些第一影音中排序在前的預設數量個第一影音產生該第一影音推荐清單。 The method according to claim 4, wherein the step of estimating the first similarity between each of the first video and audio of each of the first candidate users and the first user, and generating the first video and audio recommendation list accordingly include: Count the number of times of a first key individual of a plurality of specific keys of the historical videos; Obtaining a second key number corresponding to each specific key of each first video and audio; Estimate the first user and each first video based on the respective first key times of the specific keys of the historical videos and the second key times of each first video corresponding to each specific key The first degree of similarity; The first video and audio are sorted according to the first similarity corresponding to each of the first video and audio, and the first video and audio recommendation list is generated based on a preset number of first video and audio that are ranked first among the first videos. 如申請專利範圍第4項所述的方法，其中該第一用戶的該個人身分包括多個第一用戶資料，且所述方法包括： 取得各該既有用戶的多個參考用戶資料，其中該些參考用戶資料對應於該些第一用戶資料； 基於該第一用戶的該些第一用戶資料及各該既有用戶的該些參考用戶資料估計該第一用戶與各該既有用戶之間的一第一身分相似度； 基於各該既有用戶對應的該第一身分相似度排序該些既有用戶，並將排序在前的預設數量個既有用戶歸類至該身分分群，以作為該身分分群中的該些第一候選用戶。 The method according to claim 4, wherein the personal identity of the first user includes a plurality of first user data, and the method includes: Obtain multiple reference user data of each of the existing users, where the reference user data correspond to the first user data; Estimating a first identity similarity between the first user and each of the existing users based on the first user data of the first user and the reference user data of each of the existing users; The existing users are sorted based on the first identity similarity corresponding to each of the existing users, and the preset number of existing users ranked first are classified into the identity group as the ones in the identity group The first candidate user. 如申請專利範圍第4項所述的方法，其中依據該新影音將該第一影音推荐清單更新為該第二影音推荐清單的步驟包括： 取得該新影音的多個新關鍵子； 反應於判定該些新關鍵子的一第一新關鍵子對應於該些偏好關鍵的一第一偏好關鍵子，修正該第一偏好關鍵子的該偏好分數； 反應於判定該新影音屬於該些影音分類中的一第一影音分類，累加該第一影音分類的該歷史影音數量； 依據該第一偏好關鍵子修正後的該偏好分數及該第一影音分類累加後的該歷史影音數量重新產生該偏好分群、該觀看分群及該身分分群，其中該偏好分群、該觀看分群及該身分分群個別包括多個第二候選用戶； 取得各該第二候選用戶所觀看過的多個第二影音； 估計各該第二候選用戶的各該第二影音與該第一用戶的一第二相似度，並據以產生該第二影音推荐清單。 For the method described in item 4 of the scope of patent application, the step of updating the first audio-visual recommendation list to the second audio-visual recommendation list according to the new audio-visual recommendation list includes: Get multiple new keys for the new video; In response to determining that a first new key of the new keys corresponds to a first preference key of the preference keys, modifying the preference score of the first preference key; In response to determining that the new video and audio belong to a first video and audio category in the video and audio categories, accumulate the historical video and audio quantity of the first video and audio category; The preference group, the viewing group, and the identity group are regenerated according to the preference score after the first preference key sub-correction and the accumulated amount of the first video classification in the history, wherein the preference group, the viewing group and the identity group The identity group individually includes multiple second candidate users; Obtaining a plurality of second video and audio watched by each second candidate user; Estimating a second similarity between each second video and audio of each second candidate user and the first user, and generating the second video and audio recommendation list accordingly. 如申請專利範圍第7項所述的方法，其中依據該第一偏好關鍵子修正後的該偏好分數及該第一影音分類累加後的該歷史影音數量重新產生該偏好分群、該觀看分群及該身分分群的步驟包括： 基於該第一用戶的各該偏好關鍵子的該偏好分數與各該既有用戶對於各該偏好關鍵子的該參考偏好分數估計該第一用戶與各該既有用戶之間的一第二偏好相似度； 基於各該既有用戶對應的該第二偏好相似度排序該些既有用戶，並將排序在前的預設數量個既有用戶歸類至該偏好分群，以作為該偏好分群中的該些第二候選用戶； 基於該些歷史影音在各該影音分類中的該歷史影音數量及各該既有用戶的該些參考影音在各該影音分類中的該參考影音數量估計該第一用戶與各該既有用戶之間的一第二分類相似度； 基於各該既有用戶對應的該第二分類相似度排序該些既有用戶，並將排序在前的預設數量個既有用戶歸類至該觀看分群，以作為該觀看分群中的該些第二候選用戶。 For example, the method according to item 7 of the scope of patent application, wherein the preference group, the viewing group and the historical video quantity after the accumulation of the first video classification are regenerated according to the modified preference score of the first preference key. The steps of identity grouping include: Estimate a second preference between the first user and each existing user based on the preference score of each preference key of the first user and the reference preference score of each existing user for each preference key Similarity The existing users are sorted based on the second preference similarity corresponding to each of the existing users, and the preset number of existing users ranked first are classified into the preference group as the preference groups Second candidate user; Based on the number of historical videos in each video category of the historical videos and the number of reference videos of each existing user in each video category, it is estimated that the first user and each of the existing users A second category similarity between The existing users are sorted based on the second classification similarity corresponding to each of the existing users, and the preset number of existing users ranked first are classified into the viewing group as the viewing group. The second candidate user. 如申請專利範圍第8項所述的方法，其中估計各該第二候選用戶的各該第二影音與該第一用戶的該第二相似度，並據以產生該第二影音推荐清單的步驟包括： 取得各該第二影音對應於各該偏好關鍵子的一第二影音偏好分數； 基於各該第二影音對應於各該偏好關鍵子的該第二影音偏好分數及該第一用戶對應各該偏好關鍵子的該偏好分數估計該第一用戶與各該第二影音之間的該第二相似度； 依據各該第二影音對應的該第二相似度排序該些第二影音，並基於該些第二影音中排序在前的預設數量個第二影音產生該第二影音推荐清單。 The method according to item 8 of the scope of patent application, wherein the step of estimating the second similarity between each second video and audio of each second candidate user and the first user, and generating the second video and audio recommendation list accordingly include: Obtain a second audio-visual preference score corresponding to each of the preference keys for each second audio-visual; Based on the second audio-visual preference scores of each of the second audio-visuals corresponding to each of the preference keys and the preference scores of the first user corresponding to each of the preference keys, the estimation of the difference between the first user and each of the second audio-visuals Second similarity The second video and audio are sorted according to the second similarity corresponding to each of the second video and audio, and the second video and audio recommendation list is generated based on a preset number of second audio and video that are ranked first among the second videos. 如申請專利範圍第8項所述的方法，其中估計各該第二候選用戶的各該第二影音與該第一用戶的該第二相似度，並據以產生該第二影音推荐清單的步驟包括： 取得各該第二影音對應於各該特定關鍵子的一第三關鍵子次數； 基於該些歷史影音的該些特定關鍵子個別的該第一關鍵子次數及各該第二影音對應於各該特定關鍵子的該第三關鍵子次數估計該第一用戶與各該第二影音的該第二相似度； 依據各該第二影音對應的該第二相似度排序該些第二影音，並基於該些第二影音中排序在前的預設數量個第二影音產生該第二影音推荐清單。 The method according to item 8 of the scope of patent application, wherein the step of estimating the second similarity between each second video and audio of each second candidate user and the first user, and generating the second video and audio recommendation list accordingly include: Obtaining a third key number corresponding to each specific key of each second video and audio; Estimate the first user and each second video based on the respective first key times of the specific keys of the historical videos and the third key time of each second video corresponding to each specific key The second degree of similarity; The second video and audio are sorted according to the second similarity corresponding to each of the second video and audio, and the second video and audio recommendation list is generated based on a preset number of second audio and video that are ranked first among the second videos. 一種影音推荐系統，包括： 一前端設備，其屬於一第一用戶； 一後端設備，其經配置以： 基於該第一用戶的一偏好、一歷史觀看記錄及一個人身分從多個既有用戶中找出多個第一用戶分群，其中各該第一用戶分群包括多個第一候選用戶； 取得各該第一候選用戶所觀看過的多個第一影音； 估計各該第一候選用戶的各該第一影音與該第一用戶的一第一相似度，並據以產生一第一影音推荐清單； 反應於判定該些第一候選用戶的其中之一所觀看的一新影音不屬於該第一影音推荐清單，依據該新影音將該第一影音推荐清單更新為一第二影音推荐清單，並透過該前端設備提供予該第一用戶。 An audio-visual recommendation system, including: A front-end device, which belongs to a first user; A back-end device configured to: Finding a plurality of first user groups from a plurality of existing users based on a preference, a historical viewing record and a person identity of the first user, wherein each of the first user groups includes a plurality of first candidate users; Obtaining a plurality of first videos and audios watched by each of the first candidate users; Estimating a first similarity between each of the first video and audio of each of the first candidate users and the first user, and generating a first video and audio recommendation list accordingly; In response to determining that a new video and audio watched by one of the first candidate users does not belong to the first video and audio recommendation list, the first video and audio recommendation list is updated to a second video and audio recommendation list according to the new video and audio, and through The front-end equipment is provided to the first user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW108138696A TWI712315B (en) | 2019-10-25 | 2019-10-25 | Method and system for recommending video |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW108138696A TWI712315B (en) | 2019-10-25 | 2019-10-25 | Method and system for recommending video |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| TWI712315B true TWI712315B (en) | 2020-12-01 |
| TW202118305A TW202118305A (en) | 2021-05-01 |
Family
ID=74669879
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| TW108138696A TWI712315B (en) | 2019-10-25 | 2019-10-25 | Method and system for recommending video |
Country Status (1)
| Country | Link |
|---|---|
| TW (1) | TWI712315B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090037355A1 (en) * | 2004-12-29 | 2009-02-05 | Scott Brave | Method and Apparatus for Context-Based Content Recommendation |
| US7849092B2 (en) * | 2007-08-13 | 2010-12-07 | Yahoo! Inc. | System and method for identifying similar media objects |
| US20110153663A1 (en) * | 2009-12-21 | 2011-06-23 | At&T Intellectual Property I, L.P. | Recommendation engine using implicit feedback observations |
| US20120036523A1 (en) * | 2009-01-01 | 2012-02-09 | Orca Interactive Ltd. | Adaptive Blending of Recommendation Engines |
| US9230212B2 (en) * | 2012-02-02 | 2016-01-05 | Peel Technologies, Inc. | Content based recommendation system |
| US20160328409A1 (en) * | 2014-03-03 | 2016-11-10 | Spotify Ab | Systems, apparatuses, methods and computer-readable medium for automatically generating playlists based on taste profiles |
-
2019
- 2019-10-25 TW TW108138696A patent/TWI712315B/en active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090037355A1 (en) * | 2004-12-29 | 2009-02-05 | Scott Brave | Method and Apparatus for Context-Based Content Recommendation |
| US7849092B2 (en) * | 2007-08-13 | 2010-12-07 | Yahoo! Inc. | System and method for identifying similar media objects |
| US20120036523A1 (en) * | 2009-01-01 | 2012-02-09 | Orca Interactive Ltd. | Adaptive Blending of Recommendation Engines |
| US20110153663A1 (en) * | 2009-12-21 | 2011-06-23 | At&T Intellectual Property I, L.P. | Recommendation engine using implicit feedback observations |
| US9230212B2 (en) * | 2012-02-02 | 2016-01-05 | Peel Technologies, Inc. | Content based recommendation system |
| US20160328409A1 (en) * | 2014-03-03 | 2016-11-10 | Spotify Ab | Systems, apparatuses, methods and computer-readable medium for automatically generating playlists based on taste profiles |
Also Published As
| Publication number | Publication date |
|---|---|
| TW202118305A (en) | 2021-05-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20240202455A1 (en) | 2024-06-20 | Identifying multimedia asset similarity using blended semantic and latent feature analysis |
| US10250933B2 (en) | 2019-04-02 | Remote device activity and source metadata processor |
| US9275001B1 (en) | 2016-03-01 | Updating personal content streams based on feedback |
| US8200689B2 (en) | 2012-06-12 | Apparatus, method and computer program for content recommendation and recording medium |
| US8214318B2 (en) | 2012-07-03 | Apparatus, method and computer program for content recommendation and recording medium |
| US20130283303A1 (en) | 2013-10-24 | Apparatus and method for recommending content based on user's emotion |
| US9235574B2 (en) | 2016-01-12 | Systems and methods for providing media recommendations |
| US10380649B2 (en) | 2019-08-13 | System and method for logistic matrix factorization of implicit feedback data, and application to media environments |
| US20180103293A1 (en) | 2018-04-12 | Systems and methods for identifying non-canonical sessions |
| JP2018530847A (en) | 2018-10-18 | Video information processing for advertisement distribution |
| US20180167678A1 (en) | 2018-06-14 | Interactive media system |
| CN109246450A (en) | 2019-01-18 | A kind of video display preferentially recommender system and method based on implicit information scoring |
| WO2016172306A1 (en) | 2016-10-27 | Systems and methods for improving accuracy in media asset recommendation models |
| US9432477B2 (en) | 2016-08-30 | Identifying matching video content |
| TWI712315B (en) | 2020-12-01 | Method and system for recommending video |
| CN106462618B (en) | 2019-12-20 | Dynamic current results for a second device |
| CN110996177B (en) | 2022-04-22 | Video recommendation method, device and equipment for video-on-demand cinema |
| US12223280B2 (en) | 2025-02-11 | Identifying multimedia asset similarity using blended semantic and latent feature |
| CN110719280B (en) | 2020-11-10 | Recommendation system and method for user privacy protection based on big data |
| US11838597B1 (en) | 2023-12-05 | Systems and methods for content discovery by automatic organization of collections or rails |
| Friedrich et al. | 2017 | User preference-based probability spreading for tag-aware content recommendation |
| US20150169565A1 (en) | 2015-06-18 | Method and apparatus for classifying multimedia file |
| Bambini et al. | 2012 | Recommender systems for interactive TV |