CN109960748B - Method and system for predicting watching anchor of personal live broadcast user based on knowledge graph - Google Patents
Method and system for predicting watching anchor of personal live broadcast user based on knowledge graph Download PDFInfo
- Publication number
- CN109960748B CN109960748B CN201910133577.2A CN201910133577A CN109960748B CN 109960748 B CN109960748 B CN 109960748B CN 201910133577 A CN201910133577 A CN 201910133577A CN 109960748 B CN109960748 B CN 109960748B
- Authority
- CN
- China
- Prior art keywords
- user
- anchor
- node
- weight
- predicted
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
- H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method and a system for a personal live broadcast user to watch anchor prediction based on a knowledge graph, which comprises the steps of inquiring an anchor node adjacent to a user to be predicted in the knowledge graph, and adopting the weight between the user to be predicted and the anchor node as a first contribution weight; searching a first user set which has watched the same anchor as the user to be predicted in the knowledge graph, and finding the user with the largest number of anchors which have watched the same anchor as the user to be predicted from the first user set to serve as a second user set; obtaining a second contribution weight value of each user watching the anchor node in a second user set; extracting an anchor node with a second contribution weight in the knowledge graph, judging whether the anchor node has a first contribution weight, if so, adding the first contribution weight and the second contribution weight to be used as a final contribution weight, otherwise, directly using the second contribution weight as the final contribution weight; and extracting the anchor node corresponding to the final contribution weight value larger than the threshold value as a recommendation result.
Description
Technical Field
The invention relates to the technical field of prediction recommendation, in particular to a method for predicting the anchor watching of a personal live broadcast user based on a knowledge graph.
Background
With the popularization of mobile intelligent terminal equipment and the improvement of network bandwidth, load and other performances in recent years, the real-time interactive transmission of videos becomes possible. As a typical representative of mobile-end video, live personal broadcasting has enabled a huge transition of user roles and behaviors from traditional passive viewing to being a video content recipient and an active producer and distributor. Meanwhile, the video content on the personal live broadcast platform is converted from the traditional on-demand service static storage in the server to real-time generation, and the method has strong real-time performance and uncertainty.
Different from the traditional live broadcast video mode, the personal live broadcast interactive video mode has stronger social attribute and complex and various user behaviors. The user behavior is the most direct embodiment of the activity of the user on the live broadcast platform, and the future development trend of the platform is determined. In addition, the live broadcast platform meets the follower viscosity theory, and when a user selects to watch videos, more users select favorite anchor live broadcast videos instead of specific videos of certain types. Therefore, predicting the anchor that the user may watch in the future according to the user watching history has important value for improving the user stickiness of the whole live platform.
The current mainstream prediction mode for users to watch content is mainly based on content, history and collaborative filtering. However, the content-based video prediction method requires that the server retains all video contents and has a clear label, and is not suitable for the characteristics that the update speed of the personal live broadcast platform video which disappears along with the end of live broadcast is high, and the video type cannot be accurately obtained. The collaborative filtering-based method requires that an evaluation vector is constructed according to the evaluation of each video by the user to calculate similar users, so that video contents watched by other similar users are recommended to the user. However, videos on the personal live broadcast platform are added and removed instantly, the number of the videos changes in real time, similar users are difficult to calculate accurately, and the video prediction target on the personal live broadcast platform cannot be met.
Disclosure of Invention
In order to overcome the defect that a mainstream prediction method cannot be applied to a personal live broadcast platform due to the characteristics of huge video quantity, high updating speed, no clear user and video category labels and the like of the personal live broadcast platform, the invention provides a user watching anchor prediction method based on a user relationship knowledge graph, and the accuracy of user watching anchor prediction is improved by combining the aspects of user interest, hot anchor discovery, user relationship transmission and the like, so that recommendation is carried out according to the user interest anchor, the user adhesiveness of the personal live broadcast platform is improved, and the watching experience of a user is optimized.
Specifically, the invention provides a method for predicting the anchor watching of a personal live user based on a knowledge graph, which comprises the following steps:
step 1, extracting and integrating information of each user in a personal live broadcast platform to form a user portrait, and storing the user portrait as a node attribute in a user node of a knowledge graph;
step 2, extracting the behavior of the user watching the anchor in the personal live broadcast platform, and storing the historical behavior of the user watching the anchor as the relationship between user nodes in the knowledge graph, wherein the relationship between the entities is specifically represented by a directed edge consisting of direction and interest weight;
step 3, obtaining a user to be predicted in the personal live broadcast platform, inquiring an anchor node adjacent to the user to be predicted in the knowledge graph, and adopting an interest weight between the user to be predicted and the anchor node as a first contribution weight of the anchor node watched by the predicted user;
step 4, searching a first user set which has watched the same anchor as the user to be predicted in the knowledge graph, and finding the user with the most number of anchors which is watched by the user to be predicted from the first user set to serve as a second user set which is most similar to the watching behavior and interest of the user to be predicted;
step 5, according to the user transmission relationship, obtaining a second contribution weight value of each user watching the anchor node in the second user set;
step 6, extracting the anchor node with the second contribution weight in the knowledge graph, judging whether the anchor node has the first contribution weight, if so, adding the first contribution weight and the second contribution weight to be used as a final contribution weight, otherwise, directly using the second contribution weight as the final contribution weight;
and 7, extracting the anchor node corresponding to the final contribution weight value which is greater than the threshold value, and taking the anchor node as an anchor recommendation result of the user to be predicted.
The method for the personal live broadcast user to watch the anchor prediction based on the knowledge graph is characterized in that the step 2 comprises the step of generating the interest weight w between the user node i and the indirect adjacent user node k when interest is transferred from the user node i to the indirect adjacent user node ki,k：
Wherein wi,kRepresenting the interest weight of the user node i relative to the user node k, j is the set of user nodes directly connected with the user node k,di,jrepresenting the distance between user node i and user node j.
The method for predicting the audience of the personal live broadcast user based on the knowledge graph comprises the step 5 of obtaining a second contribution weight Px,jThe calculation method of (2) is as follows:
Px,j＝Wk,j*e-2γ
wherein Wk,jRepresenting a user VkThe edge weight, γ, between anchor j is the contribution coefficient.
The method for predicting the anchor watching of the personal live broadcast user based on the knowledge graph is characterized in that the edge weight value is obtained through the following formula:
wherein a ism,jAnd the total duration of the user i watching the anchor j in the previous M days is shown, M is the total number of users in the personal live platform, and beta is an attenuation coefficient.
The method for predicting the audience anchor watching of the personal live broadcast user based on the knowledge graph comprises the following steps of 1: country, age, language, time zone of the user.
The invention also provides a system for predicting the anchor watching of the personal live broadcast user based on the knowledge graph, which comprises the following steps:
the module 1 extracts and integrates information of each user in a personal live broadcast platform to form a user portrait, and the user portrait is stored in a user node of a knowledge graph as a node attribute;
the module 2 extracts the behavior of the user watching the anchor in the personal live broadcast platform, and stores the historical behavior of the user watching the anchor as the relationship between user nodes in the knowledge graph, wherein the relationship between the entities is specifically represented by a directed edge consisting of direction and interest weight;
a module 3, acquiring a user to be predicted in the personal live broadcast platform, inquiring an anchor node adjacent to the user to be predicted in the knowledge graph, and adopting an interest weight between the user to be predicted and the anchor node as a first contribution weight of the anchor node watched by the predicted user;
the module 4 searches a first user set which has watched the same anchor as the user to be predicted in the knowledge graph, finds the user with the most number of anchors which is watched by the user to be predicted from the first user set, and uses the user as a second user set which is most similar to the watching behavior and the interest of the user to be predicted;
the module 5 obtains a second contribution weight value of each user watching the anchor node in the second user set according to the user transfer relationship;
module 6, extracting the anchor node having the second contribution weight in the knowledge graph, and judging whether the anchor node has the first contribution weight, if so, adding the first contribution weight and the second contribution weight to be used as the final contribution weight, otherwise, directly using the second contribution weight as the final contribution weight;
and the module 7 extracts the anchor node corresponding to the final contribution weight value which is greater than the threshold value as an anchor recommendation result of the user to be predicted.
The system for the personal live broadcast user watching anchor prediction based on the knowledge graph is characterized in that the module 2 comprises an interest weight w between a user node i and a non-direct adjacent user node k when interest is transmitted from the user node i to the non-direct adjacent user node ki,k：
Wherein wi,kRepresenting the interest weight of the user node i relative to the user node k, j is a user node set directly connected with the user node k, di,jRepresenting the distance between user node i and user node j.
The system for predicting the viewing anchor of the personal live broadcast user based on the knowledge graph comprises a module 5, wherein the second contribution weight value Px,jObtained by the following formula:
Px,j＝Wk,j*e-2γ
wherein Wk,jRepresenting a user VkThe edge weight, γ, between anchor j is the contribution coefficient.
The system for the personal live broadcast user to watch the anchor prediction based on the knowledge graph is characterized in that the edge weight value is obtained through the following formula:
wherein a ism,jAnd the total duration of the user i watching the anchor j in the previous M days is shown, M is the total number of users in the personal live platform, and beta is an attenuation coefficient.
The system for predicting the audience anchor of the personal live broadcast user based on the knowledge graph comprises the following information of the user in the module 1: country, age, language, time zone of the user.
The invention has the beneficial effects that: according to the prediction method for the user to watch the anchor based on the knowledge graph, the watching weight between the user and the anchor is calculated by considering the factor that the watching interest of the user is attenuated along with time; when the user watches the anchor list, the watching interests of the user and similar users are combined, so that the accuracy and diversity of prediction are greatly improved. The invention avoids the limitation caused by the characteristics of huge number of users, no clear video anchor labels and real-time change of videos and interaction in a personal live broadcast platform in the traditional prediction mode, and solves the problem that the user watches anchor prediction in the personal live broadcast.
Drawings
FIG. 1 is a flowchart of an embodiment of a prediction method for viewing a anchor by a user of a knowledge-graph-based personal live platform according to the invention;
FIG. 2 is a flow diagram illustrating the execution of one embodiment of the user knowledge graph building process of the present invention;
fig. 3 is a flowchart illustrating an embodiment of a process for predicting user viewing anchor of a personal live platform according to the present invention.
Details of the embodiments
In order to solve the limitation that accurate user video watching prediction cannot be carried out due to the fact that the number of users is huge, anchor labels and video content labels are difficult to determine, and videos are generated and disappear and change in real time in a personal live broadcast platform, the invention provides a user anchor watching prediction method based on a knowledge graph. It mainly comprises: the method comprises two parts of user knowledge graph construction and user watching anchor prediction based on the knowledge graph.
And constructing a user knowledge graph. The method mainly comprises the steps of displaying the attributes of users and the viewing relation between the users, and mining potential information from a big data map for the user to view the anchor prediction. It comprises the following steps:
step S110: and (5) entity extraction. And extracting user information in the data set of the personal live broadcast platform, wherein the user information comprises user personal information such as the country, age, language, time zone and the like of the user to form a user portrait. The user representation is then saved as a node attribute in a user node of the knowledge-graph. The extracted objects comprise all users of the live platform, including both the anchor and the users of the watching end, and further explain that all users on the live platform can watch live and can play by themselves for live, so all user entities are extracted here.
Step S120: and (5) extracting the relation. And extracting the behavior of the user watching the anchor, and storing the historical behavior of the user watching the anchor as the relationship among the entities in a map, wherein the relationship among the entities is specifically expressed as a directed edge, the direction of the edge and the interest weight of the edge. The direction of the edge indicates the viewing active-passive relationship among users, for example, a user at a viewing end views one live broadcast of an anchor, the direction is from the user at the viewing end to the anchor, the interest weight of the edge represents the preference of the user to different anchors, the preference degree mainly represents the interest weight of the viewing user to the anchor node according to the historical viewing frequency (viewing times) of the viewing end user to the anchor in the statistical log data, and the number of times represents the interest weight of the viewing user to the anchor node.
Step S130: and carrying out user clustering mining, hot anchor discovery and user relationship transitivity by using the knowledge graph obtained in the two steps to find anchor nodes which are possibly interested by the user.
After the steps are executed, the constructed user watching behavior knowledge graph is convenient for overall control of user relations, and the user gathering effect and the user interest transfer relation are combined to be used as basic input of user watching anchor prediction.
The user viewing anchor prediction based on the knowledge graph refers to the user behavior knowledge graph constructed based on the steps, the time sequence characteristics of the user behavior are added, namely, the exponential decay model is used for updating the graph weight, and the viewing anchor list among similar users is considered, so that the accuracy and the richness of the prediction result are greatly improved. The update process is used in the recommendation process, and in the following formula, m is the previous m days of the user predicted at this time (in this embodiment, the 8 th day), for example, m is 1 in the previous day, so that the 8 th-1 th-7 th day e-1The weight of the viewing interest of (1) is large, m is 5 in the first 5 days, so 8-5 is 3 days e-3The weight of (a) is small to reflect the time-series characteristics of the user's behavior. The method specifically comprises the following steps:
step S210: for the user to be predicted, all anchor nodes associated with the user to be predicted are inquired in the knowledge graph, and the contribution weight of the user watching the anchor nodes is represented by the edge weight between the user to be predicted and the anchor nodes. The association refers to a node having a connection relationship in the graph, that is, a user node pointed by the user node to be predicted, or an anchor node having an adjacent edge of the user node to be predicted. Where a user may be both a anchor and a viewer, it is defined in the present invention that a user is an anchor node as long as the user has already been anchor, even if the anchor has already been anchor once.
Step S220: and searching a user set which has watched the same anchor as the user to be predicted, and finding the users with the most number of anchors which are watched by the user (number self-taking), namely the user set which has the most similar watching behaviors and interests to the user to be predicted.
Step S230: and calculating the viewing contribution weight of the users in the user set similar to the user to be predicted according to the user transfer relationship.
Step S240: and adding the anchor viewing contribution weight obtained in the step S210 and the viewing contribution weight of each anchor node obtained in the step S230 to obtain a final result, which is used as a final contribution weight for the user to be predicted to view the anchor.
Step S250: and sorting all the anchor viewing contribution weights in descending order from large to small, and returning the first N (different choices can be made according to specific actual requirements) anchor nodes as final predicted anchor list results.
After the steps are executed, the result of the user viewing the anchor prediction is obtained, and the processing process of the method is ended at the previous layer.
In order to make the aforementioned features and effects of the present invention more comprehensible, embodiments accompanied with figures are described in detail below.
Before explaining the embodiments, the following definitions are made for the user relationship transmission structure and the calculation of the interest weight updating method in the relationship graph (both are part of the knowledge graph), and this definition is only used to explain one embodiment of the present invention, and is not limited to the present invention and this embodiment:
(1) according to the principle that the user behaviors are locally similar, the interest of the user is attenuated layer by layer in the knowledge graph. The attenuation mode of generating the interest weight between i and k when interest is transferred from the node i to the indirect adjacent node k adopted by the method is defined as follows:
wherein wi,kRepresenting the interest weight of the user node i relative to the node k, j being the set of nodes directly connected to the node k, di,jRepresenting the distance (which may be represented in hops) between node i and node j.
(2) The invention adds the time sequence characteristics of the user behavior when predicting that the user watches the anchor list, thereby obtainingMore accurate prediction results are obtained. The method adopts an exponential decay model to correct interest weight information in the knowledge graph. Assuming that the first m days of the user's viewing history is used as the source of the predicted data of interest weight, the viewing weight (edge weight) W between the ith user and the jth anchori,jThe definition is as follows:
wherein a ism,jThe total duration of the user i watching the anchor j on the previous M days means that the total duration of the watching on the current M days of the current day is predicted, for example, the eighth day is predicted, M is 2, that is, the total duration of the watching on the 6 th day, β is a specified attenuation coefficient (which may take a value of 1), and M is the total number of users in the personal live platform.
In order to avoid the problem of prediction limitation caused by the characteristics of huge quantity of users and video contents, unclear video and anchor labels, interactivity and the like on a personal live broadcast platform by a traditional prediction method, the invention provides a user anchor watching prediction method based on a knowledge graph, which solves the problem of user anchor watching prediction on the personal live broadcast platform under the condition that direct user interests such as the anchor labels and video category labels do not need to be obtained, and improves the activity and the adhesiveness of platform users. Fig. 1 is a flowchart of an embodiment of a method for predicting a user viewing anchor based on a knowledge graph, which mainly includes: the method comprises two parts of user knowledge graph construction and user watching anchor prediction based on the knowledge graph.
The user knowledge graph is constructed, the viewing relation between the user attribute and the user is clearly shown in a graph mode, and potential information is mined from the graph, such as user local clustering, hot anchor discovery, user interest transfer relation and the like, and is used for user viewing anchor prediction. As shown in fig. 2, the process mainly includes the following steps:
step S110: and (5) entity extraction. And selecting user log data of the first eight days of the live broadcast platform, and extracting user anchor information in a personal live broadcast platform data set, wherein the user anchor information comprises user personal information such as the country, age, language, time zone and the like of a user to form a user portrait. The user representation is then saved as a node attribute in a user node of the knowledge-graph. The user node information obtained by entity extraction includes user id corresponding to the user node, user birthday, country where the user node is located, number of concerned users, number of owned fans, gender and the like.
Step S120: and (5) extracting the relation. Extracting the relation among different users according to the behavior of watching the anchor in the last eight days of the users, generating the contact edges among the users according to the watching relation, wherein the edges point to the watched anchor nodes by the viewers, and the weights of the edges are obtained by calculating the watching frequency of the edges.
Step S130: and (3) carrying out user cluster mining, hot anchor discovery and user relationship transitivity to find anchor nodes which are possibly interested by the user by using the knowledge graph obtained in the two steps S110 and S120. The method comprises the following steps: the generated user viewing behavior knowledge graph is stored in a graph database Neo4j, and the user category formed by clustering is automatically obtained according to user density distribution; then, according to the information of the user vermicelli amount and the watched times, hot anchor nodes (the nodes at the front are arranged in descending order according to the degree of income) are found out; and finally, searching for the anchor node which is possibly interested by the user through the relationship transfer effect among the users in the graph. In the embodiment, the interest attenuation formula in definition (1) is adopted, and the calculation complexity is considered, and the weight w of interest transfer between all user nodes i in the graph and non-adjacent nodes k is calculated and obtained by using three-layer user transfer (three-step hop count)i,k。
After the steps are executed, the constructed user watching behavior knowledge graph is convenient for overall control of user relations, and the user gathering effect and the user interest transfer relation are combined to be used as basic input of user watching anchor prediction.
The user viewing anchor prediction based on the knowledge graph refers to the user behavior knowledge graph constructed based on the steps, the time sequence characteristics of the user behavior are added, namely, the exponential decay model is used for updating the graph weight, and the viewing anchor list among similar users is considered, so that the accuracy and the richness of the prediction result are greatly improved. Firstly, extracting the data of the first eight days to obtain 8 groups of characteristics, and calculating the influence of the characteristics obtained by the data of each day on the viewing behaviors of the ninth day to obtain the importance degree of the viewing behaviors at different times on the prediction result, namely the importance degree is in negative correlation with the distance from the time point to be predicted. As shown in fig. 3, the method specifically includes the following steps:
step S210: for a user x to be predicted, all anchor nodes associated with the user x to be predicted are inquired in the knowledge graph, and the contribution weight of the anchor node watched by the user x is represented by the edge weight between the user x to be predicted and the anchor node. The contribution weight here refers to the weight that x points to each of the anchor nodes. For example, if the user to be predicted watches 2 times the main broadcast y, then the contribution value is 2.
Step S220: searching for users who have watched the same anchor as the user x to be predicted in the knowledge graph, and finding out K users (V) with the maximum number of anchors same as the user x in watching1,V2,…,VK) I.e. the set of users that most closely resembles the viewing behavior of user x to be predicted. All the anchor watched by the user x to be predicted, namely the node pointed by x in the directed graph, can be seen in the knowledge graph, and a watching set can be formed actually, then the anchor set is watched by other users in history, and the user with the largest number of intersections is the user most similar to the user x.
Step S230: and calculating the viewing contribution weight of the users in the user set similar to the user x to be predicted according to the user transfer relationship. For user V in the similar set (second user set)kThe user x to be predicted can be watched in the future due to similar interests of the anchor node j in historical watching, so the watching contribution value P is calculatedx,jAs follows below, the following description will be given,
Px,j＝Wk,j*e-2γ
wherein Wk,jRepresenting a user VkThe edge weight, γ, between anchor j is the contribution coefficient. The contribution factor is the number of hops from x to j in the knowledge-graph, e.g. A->B,B->C then A also points indirectly to C, so if A->The contribution coefficient of B is 1, A->The C contribution factor is 2.
Step S240: the final result of adding the anchor viewing contribution weight calculated in step S210 and the viewing contribution weight of each anchor node calculated in step S230 is that, as the final contribution weight of user x viewing the anchor, for example, some anchor nodes will have only the viewing contribution value calculated in step S230 because they are not viewed by the user to be predicted, and the final contribution value of this part of anchor nodes is the viewing contribution value calculated in step S230 because their anchor viewing contribution weight calculated in step S210 is 0. The likelihood of the user to be predicted (the prediction objective of the present invention is which anchor the user to be predicted will view) depends not only on his own interest in the anchor, i.e. the contribution value calculated in S210, but also on the viewing contribution value similar to him (the user who may have the same interest), i.e. the contribution value calculated in S230. The final value is thus the sum of the two contribution values, which together represent the likelihood, or interest in the anchor, of the user to be predicted to watch the anchor.
Step S250: and sorting all the anchor viewing contribution weights in descending order from large to small, and returning the first N anchor nodes as final predicted anchor list results.
After the steps are executed, the result of the user viewing the anchor prediction is obtained, and the processing process of the method is ended at the previous layer.
The following are system examples corresponding to the above method examples, and this embodiment can be implemented in cooperation with the above embodiments. The related technical details mentioned in the above embodiments are still valid in this embodiment, and are not described herein again in order to reduce repetition. Accordingly, the related-art details mentioned in the present embodiment can also be applied to the above-described embodiments.
The invention also provides a system for predicting the anchor watching of the personal live broadcast user based on the knowledge graph, which comprises the following steps:
the module 1 extracts and integrates information of each user in a personal live broadcast platform to form a user portrait, and the user portrait is stored in a user node of a knowledge graph as a node attribute;
the module 2 extracts the behavior of the user watching the anchor in the personal live broadcast platform, and stores the historical behavior of the user watching the anchor as the relationship between user nodes in the knowledge graph, wherein the relationship between the entities is specifically represented by a directed edge consisting of direction and interest weight;
a module 3, acquiring a user to be predicted in the personal live broadcast platform, inquiring an anchor node adjacent to the user to be predicted in the knowledge graph, and adopting an interest weight between the user to be predicted and the anchor node as a first contribution weight of the anchor node watched by the predicted user;
the module 4 searches a first user set which has watched the same anchor as the user to be predicted in the knowledge graph, finds the user with the most number of anchors which is watched by the user to be predicted from the first user set, and uses the user as a second user set which is most similar to the watching behavior and the interest of the user to be predicted;
the module 5 obtains a second contribution weight value of each user watching the anchor node in the second user set according to the user transfer relationship;
module 6, extracting the anchor node having the second contribution weight in the knowledge graph, and judging whether the anchor node has the first contribution weight, if so, adding the first contribution weight and the second contribution weight to be used as the final contribution weight, otherwise, directly using the second contribution weight as the final contribution weight;
and the module 7 extracts the anchor node corresponding to the final contribution weight value which is greater than the threshold value as an anchor recommendation result of the user to be predicted.
The system for the personal live broadcast user watching anchor prediction based on the knowledge graph is characterized in that the module 2 comprises an interest weight w between a user node i and a non-direct adjacent user node k when interest is transmitted from the user node i to the non-direct adjacent user node ki,k：
Wherein wi,kRepresenting interest weight of user node i relative to user node k, j being directly connected to user node kSet of user nodes of di,jRepresenting the distance between user node i and user node j.
The system for predicting the viewing anchor of the personal live broadcast user based on the knowledge graph comprises a module 5, wherein the second contribution weight value Px,jObtained by the following formula:
Px,j＝Wk,j*e-2γ
wherein Wk,jRepresenting a user VkThe edge weight, γ, between anchor j is the contribution coefficient.
The system for the personal live broadcast user to watch the anchor prediction based on the knowledge graph is characterized in that the edge weight value is obtained through the following formula:
wherein a ism,jAnd the total duration of the user i watching the anchor j in the previous M days is shown, M is the total number of users in the personal live platform, and beta is an attenuation coefficient.
The system for predicting the audience anchor of the personal live broadcast user based on the knowledge graph comprises the following information of the user in the module 1: country, age, language, time zone of the user.
Claims (10)
1. A method for predicting the audience anchor of a personal live broadcast user based on a knowledge graph is characterized by comprising the following steps:
step 1, extracting and integrating information of each user in a personal live broadcast platform to form a user portrait, and storing the user portrait as a node attribute in a user node of a knowledge graph;
step 2, extracting the behavior of the user watching the anchor in the personal live broadcast platform, and storing the historical behavior of the user watching the anchor as the relationship among user nodes in the knowledge graph, wherein the relationship is embodied as a directed edge consisting of direction and interest weight;
step 3, obtaining a user to be predicted in the personal live broadcast platform, inquiring an anchor node adjacent to the user to be predicted in the knowledge graph, and adopting an interest weight between the user to be predicted and the anchor node as a first contribution weight of the anchor node watched by the predicted user;
step 4, searching a first user set which has watched the same anchor as the user to be predicted in the knowledge graph, and finding the user with the most number of anchors which is watched by the user to be predicted from the first user set to serve as a second user set which is most similar to the watching behavior and interest of the user to be predicted;
step 5, according to the user transmission relationship, obtaining a second contribution weight value of each user watching the anchor node in the second user set;
step 6, extracting the anchor node with the second contribution weight in the knowledge graph, judging whether the anchor node has the first contribution weight, if so, adding the first contribution weight and the second contribution weight to be used as the final contribution weight, otherwise, directly using the second contribution weight as the final contribution weight;
and 7, extracting the anchor node corresponding to the final contribution weight value which is greater than the threshold value, and taking the anchor node as an anchor recommendation result of the user to be predicted.
2. The method of claim 1, wherein step 2 comprises generating the interest weight w between i and k when interest is transferred from user node i to non-directly adjacent user node ki,k：
Wherein wi,kRepresenting the interest weight of the user node i relative to the user node k, j is a user node set directly connected with the user node k, di,jRepresenting the distance between user node i and user node j.
3. A method of knowledge-graph based personal live user watch anchor prediction according to claim 1 or 2,wherein the second contribution weight P in step 5x,jThe calculation method of (2) is as follows:
wherein Wk,jRepresenting a user VkThe edge weight, γ, between anchor j is the contribution coefficient.
4. The method of claim 3, wherein the edge weights are derived by:
wherein a ism,jAnd the total duration of the user i watching the anchor j in the previous M days is shown, M is the total number of users in the personal live platform, and beta is an attenuation coefficient.
5. The method of knowledge-graph-based personal live user watch anchor prediction as claimed in claim 4, wherein the user's information in step 1 comprises: country, age, language, time zone of the user.
6. A system for knowledge-graph based on a prediction of a cast watched by a live personal user, comprising:
the module 1 extracts and integrates information of each user in a personal live broadcast platform to form a user portrait, and the user portrait is stored in a user node of a knowledge graph as a node attribute;
the module 2 extracts the behavior of the user watching the anchor in the personal live broadcast platform, and stores the historical behavior of the user watching the anchor as the relationship among user nodes in the knowledge graph, wherein the relationship is embodied as a directed edge consisting of direction and interest weight;
a module 3, acquiring a user to be predicted in the personal live broadcast platform, inquiring an anchor node adjacent to the user to be predicted in the knowledge graph, and adopting an interest weight between the user to be predicted and the anchor node as a first contribution weight of the anchor node watched by the predicted user;
the module 4 searches a first user set which has watched the same anchor as the user to be predicted in the knowledge graph, finds the user with the most number of anchors which is watched by the user to be predicted from the first user set, and uses the user as a second user set which is most similar to the watching behavior and the interest of the user to be predicted;
the module 5 obtains a second contribution weight value of each user watching the anchor node in the second user set according to the user transfer relationship;
module 6, extracting the anchor node having the second contribution weight in the knowledge graph, and determining whether the anchor node has the first contribution weight, if so, adding the first contribution weight and the second contribution weight to be used as the final contribution weight, otherwise, directly using the second contribution weight as the final contribution weight;
and the module 7 extracts the anchor node corresponding to the final contribution weight value which is greater than the threshold value as an anchor recommendation result of the user to be predicted.
7. The system of claim 6, wherein the module 2 comprises generating the interest weight w between i and k when interest is transferred from a user node i to a non-directly adjacent user node ki,k：
Wherein wi,kRepresenting the interest weight of the user node i relative to the user node k, j is a user node set directly connected with the user node k, di,jRepresenting the distance between user node i and user node j.
8. As in claimThe system for knowledge-graph-based on-demand live user watch-cast prediction of a personal live user according to claim 6 or 7, characterized in that the second contribution weight P in the module 5x,jObtained by the following formula:
wherein Wk,jRepresenting a user VkThe edge weight, γ, between anchor j is the contribution coefficient.
9. The system of claim 8, wherein the edge weights are derived by:
wherein a ism,jAnd the total duration of the user i watching the anchor j in the previous M days is shown, M is the total number of users in the personal live platform, and beta is an attenuation coefficient.
10. The system for knowledge-graph based on live personal user watch anchor prediction as claimed in claim 9, wherein the user's information in module 1 includes: country, age, language, time zone of the user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910133577.2A CN109960748B (en) | 2019-02-22 | 2019-02-22 | Method and system for predicting watching anchor of personal live broadcast user based on knowledge graph |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910133577.2A CN109960748B (en) | 2019-02-22 | 2019-02-22 | Method and system for predicting watching anchor of personal live broadcast user based on knowledge graph |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109960748A CN109960748A (en) | 2019-07-02 |
| CN109960748B true CN109960748B (en) | 2021-03-26 |
Family
ID=67023659
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910133577.2A Active CN109960748B (en) | 2019-02-22 | 2019-02-22 | Method and system for predicting watching anchor of personal live broadcast user based on knowledge graph |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109960748B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110825915B (en) * | 2019-11-12 | 2022-05-13 | 深圳墨世科技有限公司 | Short video label extraction method and device, computer equipment and storage medium |
| CN111737594B (en) * | 2020-06-24 | 2023-07-25 | 中网数据(北京)股份有限公司 | Virtual network role behavior modeling method based on unsupervised label generation |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106095858A (en) * | 2016-06-02 | 2016-11-09 | 海信集团有限公司 | A kind of audio video searching method, device and terminal |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10120877B2 (en) * | 2011-09-15 | 2018-11-06 | Stephan HEATH | Broad and alternative category clustering of the same, similar or different categories in social/geo/promo link promotional data sets for end user display of interactive ad links, coupons, mobile coupons, promotions and sale of products, goods and services integrated with 3D spatial geomapping and mobile mapping and social networking |
| US9270714B2 (en) * | 2014-03-13 | 2016-02-23 | International Business Machines Corporation | Content preview generation using social network analysis |
| US9762934B2 (en) * | 2014-11-04 | 2017-09-12 | Electronics And Telecommunications Research Institute | Apparatus and method for verifying broadcast content object identification based on web data |
| CN107122399A (en) * | 2017-03-16 | 2017-09-01 | 中国科学院自动化研究所 | Combined recommendation system based on Public Culture knowledge mapping platform |
| CN108897887B (en) * | 2018-07-10 | 2020-10-16 | 华南师范大学 | Teaching resource recommendation method based on knowledge graph and user similarity |
-
2019
- 2019-02-22 CN CN201910133577.2A patent/CN109960748B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106095858A (en) * | 2016-06-02 | 2016-11-09 | 海信集团有限公司 | A kind of audio video searching method, device and terminal |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109960748A (en) | 2019-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20210326674A1 (en) | Content recommendation method and apparatus, device, and storage medium | |
| CN111708901B (en) | Multimedia resource recommendation method and device, electronic equipment and storage medium | |
| EP3819791A2 (en) | Information search method and apparatus, device and storage medium | |
| CN107038213B (en) | Video recommendation method and device | |
| US9955231B2 (en) | Relevant video content pushed to a mobile phone | |
| CN107968952B (en) | Method, device, server and computer-readable storage medium for recommending videos | |
| US8069076B2 (en) | Generating audience analytics | |
| CN103974097B (en) | Personalized user original video forecasting method based on popularity and social networkies and system | |
| CN110430471A (en) | It is a kind of based on the television recommendations method and system instantaneously calculated | |
| JP4370850B2 (en) | Information processing apparatus and method, program, and recording medium | |
| US20090100094A1 (en) | Recommendation system and method for multimedia content | |
| CN102056018B (en) | Method and system for providing TV guide and method for providing program-requesting information | |
| CN107818166A (en) | A kind of information recommends method, apparatus, server and system | |
| CN113301442B (en) | Method, device, medium, and program product for determining live broadcast resource | |
| JP2008542870A (en) | Method and apparatus for estimating the overall interest of a group of users for content | |
| US9672534B2 (en) | Preparing content packages | |
| CN103997662A (en) | Program pushing method and system | |
| CN109960748B (en) | Method and system for predicting watching anchor of personal live broadcast user based on knowledge graph | |
| CN112579913A (en) | Video recommendation method, device, equipment and computer-readable storage medium | |
| CN111552835B (en) | File recommendation method, device and server | |
| JP6823170B2 (en) | Prediction of interruptions in the content stream | |
| CN110996177B (en) | Video recommendation method, device and equipment for video-on-demand cinema | |
| CN112785328A (en) | Content pushing method and device and computer storage medium | |
| KR20110071452A (en) | Broadcast program recommendation method and apparatus using genres weight | |
| Papadakis et al. | Movie score: Personalized movie recommendation on mobile devices |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |