CN115168744A

CN115168744A - Radio and television technology knowledge recommendation method based on user portrait and knowledge graph

Info

Publication number: CN115168744A
Application number: CN202210947035.0A
Authority: CN
Inventors: 江立宇; 刘飞; 尤浩东
Original assignee: Zhoushan Radio And Television Station
Current assignee: Zhoushan Radio And Television Station
Priority date: 2022-08-09
Filing date: 2022-08-09
Publication date: 2022-10-11

Abstract

The invention provides a broadcasting and television technology knowledge recommendation method based on user portrait and a knowledge map, and relates to the technical field of intelligent broadcasting and television. On one hand, a set of user portrait of broadcasting and television technology knowledge is constructed from multiple dimensions, and the similarity among users is calculated by combining a collaborative filtering algorithm, so that the items which are interested by the users are predicted based on favorite items of similar users; on the other hand, a recommendation model based on the knowledge graph is constructed, potential preferences of the user are mined spontaneously through preference diffusion, knowledge graph feature learning is merged into the recommendation model based on the recurrent neural network, and deeper user preferences are obtained, so that items which the user is interested in are predicted. And finally, sequencing the interested items obtained in the two modes according to the scores, extracting the first several items and pushing the items to the user. By combining two modes of knowledge maps and user images, the intelligent level recommended by the radio and television knowledge intelligence base can be effectively improved, and the recommendation result is more accurate.

Description

Radio and television technology knowledge recommendation method based on user portrait and knowledge graph

Technical Field

The invention relates to the technical field of intelligent broadcasting and television, in particular to a broadcasting and television technology knowledge recommendation method based on user portrait and knowledge graph.

Background

At present, the digital reform is in the important stage of 'comprehensive through, integrated breakthrough and centralized display', and is in the era of digital reform, and the city and county level television grandma of each province takes the digital reform as an important political task and an important hand grip for promoting the innovation. In order to better promote a new round of high-quality development of city and county level local television stations in the media-integrated era, the intelligent recommendation system aims to establish a radio and television technology knowledge base by taking digital reformation as a guide. The recommendation system can help users to efficiently screen out interesting information, and is beneficial to television audiences to conveniently and effectively obtain favorite program contents. The recommendation system aims to establish a user model and an article model by utilizing the collected user information and the target article information, match according to close-up rules, filter calculation results by utilizing the rules contained in the algorithm, find out commodities which are possibly interested by the user and recommend the commodities to the user.

However, the traditional recommendation method, such as collaborative filtering, depends on historical preference data of users, and the preference data is very sparse, so that the recommendation accuracy is reduced, and meanwhile, new users and new articles still face the cold start problem. Knowledge Graph (KG) is proposed by Google in 2012, is an emerging auxiliary information in recent years, expresses relation dependence among different knowledge nodes and the like by means of characteristics of graph structures, can integrate and extract knowledge from multi-source heterogeneous data, but the existing knowledge graph is not further mined by combining feature information, lacks some hidden features, and leads to inaccurate recommendation results.

Disclosure of Invention

The invention aims to provide a broadcasting and television technology knowledge recommendation method based on user portrait and knowledge graph, which improves the intelligent level of broadcasting and television knowledge intelligence recommendation by sufficiently organizing and modeling a broadcasting and television technology knowledge base and using the knowledge graph and a user image mode, and the recommendation result is more accurate.

The embodiment of the invention is realized by the following steps:

in a first aspect, an embodiment of the present application provides a radio and television technology knowledge recommendation method based on a user portrait and a knowledge graph, which includes:

acquiring basic attribute information and behavior characteristic information of a user, and quantizing, fusing and reconstructing the information to form a user characteristic vector;

based on the user feature vectors, calculating by using a collaborative filtering algorithm to obtain similarity results among users, and performing collaborative recommendation according to the similarity results to obtain a first recommended neighbor set;

acquiring historical preference data of a user, and performing diffusion propagation in a pre-constructed knowledge graph in the field of broadcasting and television according to the historical preference data to obtain an integral diffusion preference set of the user;

inputting the user preference vector in the overall diffusion preference set into a preset RNN recommendation model for prediction to obtain a second recommendation neighbor set;

and fusing the first recommended neighbor set and the second recommended neighbor set to obtain a final recommended set, and pushing results in the recommended set to the user.

Based on the first aspect, in some embodiments of the present invention, the step of obtaining the basic attribute information and the behavior feature information of the user, and performing quantization and fusion reconstruction on the information to form the user feature vector includes:

encoding the basic attribute information of the user by utilizing a one-hot encoding mode to obtain a basic attribute vector of the user;

extracting keywords from the behavior characteristic information of the user by using a TF-IDF algorithm, and converting the keywords into a vector form to obtain a user behavior vector;

and forming a user characteristic vector based on the user basic attribute vector and the user behavior vector.

Based on the first aspect, in some embodiments of the present invention, the calculating, based on the user feature vector, a similarity result between the users by using a collaborative filtering algorithm, and performing collaborative recommendation according to the similarity result to obtain the first recommended neighbor set includes:

according to the basic attribute vector and the user behavior vector of the user, respectively calculating by using a cosine similarity algorithm to obtain attribute similarity and behavior similarity between the users;

fusing the attribute similarity and the behavior similarity according to a fusion formula to obtain a similarity result between the users so as to obtain similar users corresponding to the users through matching;

and acquiring a radio and television content item table corresponding to the similar user and recommending the radio and television content item table to the user to form a first recommended neighbor set.

In some embodiments of the invention based on the first aspect, the fusion formula is:

where SIM (p, q) represents the similarity result between user p and user q, SIM _user (p, q) denotes attribute similarity between subscribers, SIM _tech (p, q) represents the behavior similarity between users, and x is a weight coefficient and represents the ratio of the number of the same attributes between users to the vector dimension of the basic attribute of the users.

Based on the first aspect, in some embodiments of the present invention, the method further includes:

acquiring radio and television field knowledge data, and extracting information to obtain entity information, relationship information and attribute information of an entity to form a knowledge triple;

and (3) carrying out feature learning by using a TransE algorithm based on the knowledge triples to obtain entities and relationship vectors and form a knowledge map in the field of broadcasting and television.

Based on the first aspect, in some embodiments of the present invention, the step of performing diffusion propagation in a pre-constructed knowledge graph in the field of radio and television according to historical preference data to obtain an overall diffusion preference set of the user includes:

traversing the historical preferences of the user, and combining to form a historical preference set of the user;

taking each entity node in the historical preference set as a starting point, connecting to a related entity corresponding to the entity through a knowledge graph, and counting to form a first-layer diffusion preference set;

taking each entity node in the first layer of diffusion preference set as a starting point, connecting to a corresponding associated entity of the entity through a knowledge graph, and counting to form a second layer of diffusion preference set;

and fusing the historical preference set, the first-layer diffusion preference set and the second-layer diffusion preference set to obtain an overall diffusion preference set of the user.

Based on the first aspect, in some embodiments of the present invention, the RNN recommendation model employs an item-based attention mechanism, and performs linear combination on different input parts according to a hierarchical relationship of a user diffusion preference set to form a preference feature representation of the user, so as to predict a preference item of the user, and train the model by minimizing cross entropy loss between a true value and a predicted value.

Based on the first aspect, in some embodiments of the present invention, the step of fusing the first recommended neighbor set and the second recommended neighbor set to obtain a final recommended set, and pushing results in the recommended set to the user includes:

respectively sorting the recommendation results in the first recommendation neighbor set and the second recommendation neighbor set according to scores;

extracting a plurality of previous recommendation results from the first recommendation neighbor set and the second recommendation neighbor set to form a push list;

and after the recommendation results in the push list are screened and cleaned, pushing the recommendation results to the user.

In a second aspect, an embodiment of the present application provides an electronic device, which includes a memory for storing one or more programs; a processor. The one or more programs, when executed by the processor, implement the method as described in any of the first aspects above.

In a third aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements the method as described in any one of the above first aspects.

Compared with the prior art, the embodiment of the invention has at least the following advantages or beneficial effects:

on one hand, a set of radio and television technology knowledge user portrait is constructed from multiple dimensions based on the basic attributes and dynamic behavior characteristics of users, the similarity among the users is calculated by combining a collaborative filtering algorithm after the information of each dimension of the users is quantized, fused and reconstructed, and thus the items which are interested by the users are predicted based on the favorite items of the similar users, and the personalized recommendation of the radio and television technology knowledge is realized; on the other hand, a recommendation model based on the knowledge graph is constructed, potential preferences of the user are mined spontaneously through preference diffusion, knowledge graph features are learned and merged into the recommendation model, and deeper user preferences are obtained through a recurrent neural network, so that items which are interested by the user are predicted. And finally, the items which are obtained in the two ways and are possibly interested by the user are ranked according to the scores, a plurality of items are taken to form a recommendation list, and the recommendation list is pushed to the user. This application organizes and models through abundant broadcasting and television technology knowledge base, utilizes the mode of knowledge map and user's image, can effectively promote the intelligent level that the broadcasting and television knowledge intelligence base was recommended for the recommendation result is more accurate, improves user satisfaction. The scheme provided by the application can be further applied to broadcasting and television and related industries, is used for building, spreading and culturing the knowledge base, can effectively improve the digitization technology and application level of the broadcasting and television industry, and promotes effective promotion of data reformation work.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.

FIG. 1 is a flow chart of an embodiment of a broadcasting and television technology knowledge recommendation method based on a user portrait and a knowledge graph according to the present invention;

fig. 2 is a schematic flow chart of a step of calculating a similarity result between users by using a collaborative filtering algorithm based on a user feature vector in an embodiment of a radio and television technology knowledge recommendation method based on a user portrait and a knowledge graph, and performing collaborative recommendation according to the similarity result to obtain a first recommended neighbor set;

FIG. 3 is a schematic diagram of knowledge graph connection in an embodiment of a radio and television technology knowledge recommendation method based on a user portrait and a knowledge graph according to the present invention;

fig. 4 is a block diagram of an electronic device according to an embodiment of the present invention.

Icon: 1. a memory; 2. a processor; 3. a communication interface.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. The components of the embodiments of the present application, generally described and illustrated in the figures herein, can be arranged and designed in a wide variety of different configurations.

Thus, the following detailed description of the embodiments of the present application, presented in the accompanying drawings, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

Examples

Some embodiments of the present application will be described in detail below with reference to the accompanying drawings. The embodiments described below and the individual features of the embodiments can be combined with one another without conflict.

Referring to fig. 1 to 3, an embodiment of the present application provides a radio and television technology knowledge recommendation method based on a user portrait and a knowledge graph, including the following steps:

step S1: and acquiring basic attribute information and behavior characteristic information of the user, and quantizing, fusing and reconstructing the information to form a user characteristic vector.

In the steps, when the user portrait is constructed, the basic attribute information of the user is coded in a one-hot coding mode from two dimensions of the static basic attribute and the dynamic behavior characteristic of the user to obtain a basic attribute vector of the user, the behavior characteristic information of the user is subjected to keyword extraction by using a TF-IDF algorithm, the keywords are converted into a vector form to obtain a behavior vector of the user, and then the two are combined to form a user characteristic vector for calculating the similarity between the users.

Specifically, the static basic attributes of the user include, but are not limited to: age, academic calendar, specialty, working years, post, technical speciality, technical style, etc. In addition, the total dimensions of the static basic attributes of the user are not many, so that the basic attribute information of the user can be encoded in a "one-hot" manner to obtain a binary vector representation. The method is to use N bit state registers to code N states, each state has its independent register bit, and each bit has only '0' and '1' states, thereby effectively improving the reading speed of information.

The dynamic behavior characteristics of the user mainly comprise the conditions of user retrieval, collection, downloading and resource reference, and are formed by user behavior vectors, and the user behavior vectors represent the preference degree of the user for a certain item or a certain type of programs. The dynamic behavior characteristics of the user are often closely related to the knowledge of the technical documents accessed by the technical user, however, a complete technical document contains too much information, and a concise vector is difficult to form for calculation. Therefore, the keyword extraction can be carried out by a TF-IDF (inverse document frequency) method, and the principle is to comprehensively judge the position of a word in a document by calculating the TF value of the word frequency of the word in the text and the IDF value of the inverse document frequency of the word. Wherein the TF value is defined as: the ratio of the occurrence frequency of the entries in a certain type of document to the number of all the entries in the certain type of document; the IDF value is defined as: log (total technical document number/(document number containing an entry + 1)). The purpose of extracting the feature words from the technical documents visited by the user is achieved through the TF-IDF algorithm, so that the dynamic behavior features of the user are obtained, and are converted into a simple vector form to obtain the user behavior vectors for calculating the similarity between the users.

Step S2: based on the user feature vectors, calculating by using a collaborative filtering algorithm to obtain similarity results between users, and performing collaborative recommendation according to the similarity results to obtain a first recommended neighbor set.

After the basic attribute vector and the user behavior vector of the user are obtained, the similarity between the users is calculated according to the basic attribute vector and the user behavior vector of the user (the user behavior vector represents the preference degree of the user on a certain project or a certain type of program) by combining a collaborative filtering algorithm, a 'neighbor' user group similar to the current user attribute (background, character and the like) and behavior (preference on the project) is found, namely the similar user group of the user is found, then recommendation is carried out on the current user based on the historical preference information of the similar user, namely collaborative recommendation is carried out according to the similarity result, and therefore the first recommended neighbor set of the user is obtained. The method specifically comprises the following steps:

step S2-1: according to the basic attribute vector and the user behavior vector of the user, respectively calculating attribute similarity and behavior similarity between the users by using a cosine similarity algorithm;

in the above steps, the formula for calculating the attribute similarity between users is:

wherein, P and Q are the basic attribute vectors of user P and user Q respectively, P _i And Q _i The basic attribute values of the dimensions of the user p and the user q, such as the dimensions of a post, a technical specialty, a technical style and the like, are respectively considered, the similarity of the static basic attribute of the user is considered from multiple aspects, and the method is favorable for finding similar users closer to the user.

The calculation formula of the behavior similarity between the users is as follows:

5,T is the behavior vectors of user p and user q, respectively, and N () represents the number of elements of the behavior feature in the set. In addition, in order to simplify the calculation, the behavior similarity of the dynamic behavior characteristics of the users adopts an inverted matrix mode, namely, firstly, an inverted list from an item to the user is experienced, namely, each item stores a user list interested in the item, and then, the similarity between the users is calculated.

Step S2-2: fusing the attribute similarity and the behavior similarity according to a fusion formula to obtain a similarity result between users so as to obtain similar users corresponding to the users through matching;

in the above steps, after obtaining the attribute similarity and behavior similarity between users, the similarity between two users is calculated according to the following formula:

where SIM (p, q) represents the similarity result between user p and user q, SIM _user (p, q) denotes attribute similarity between subscribers, SIM _tech (p, q) represents the similarity of behaviors between users, x is a weight coefficient representing the ratio of the number of identical attributes between users to the vector dimension of the basic attribute of the user,

representing SIM by sigmod function _tech And (5) carrying out a normalization result. According to the similarity result between the users, the similar users corresponding to the current user can be matched.

Step S2-3: and acquiring a broadcasting and television content item table corresponding to the similar user and recommending the broadcasting and television content item table to the user to form a first recommended neighbor set.

In the above steps, according to the matched similar user corresponding to the current user, preference items of the similar user, such as broadcast and television content items, are recommended to the current user, that is, collaborative recommendation of broadcast and television technology content is performed, so that a first recommendation proximity set is obtained.

And step S3: and acquiring historical preference data of the user, and performing diffusion propagation in a pre-constructed knowledge graph in the field of broadcasting and television according to the historical preference data to obtain an integral diffusion preference set of the user.

In the above steps, the historical access items of the user in the knowledge graph are taken as preference starting points, the user preferences are propagated through the relationship links among the entities in the knowledge graph, the potential preferences of the user are learned, and the potential preferences are used for fusing the preference features of all propagation levels by using an attention network mechanism subsequently to construct a final user preference vector, so that the probability that the user likes a certain item is predicted.

For example, referring to fig. 3, the knowledge graph is a relationship network obtained by connecting all the Heterogeneous Information (relationships) together. For example, when a user wants to recommend a movie which the user may be interested in, the user may be propagated in the knowledge graph in the radio and television field based on a certain movie which the user has historically watched, and the potential preference of the user is mined by spreading from aspects such as director, actors, and the type (tag) of the movie, so that the user is recommended more personally, and the accuracy of the recommendation and the satisfaction of the user are improved.

Specifically, knowledge data in the field of broadcasting and television are obtained, and entity information, relationship information and attribute information of an entity are obtained through information extraction to form a knowledge triple; and then, based on the knowledge triples, performing feature learning by using a TransE algorithm to obtain entities and relationship vectors, and forming a knowledge graph in the field of broadcasting and television. By adopting a distance-based translation model method (TransE algorithm), the distance between two entities with a connection relation is enabled to be as small as possible, and then the vector representation of the knowledge entity is learned. In the training process, a sigmod function is adopted as a related loss function, and the obtained normalized result is as follows:

where (h, r, t) represents a knowledge-based triplet in the knowledge-graph.

After the knowledge graph in the broadcasting and television field is obtained through the TransE algorithm learning, the diffusion propagation can be carried out in the constructed knowledge graph in the broadcasting and television field according to the historical preference data of the current user, and the whole diffusion preference set of the current user is obtained.

Specifically, firstly, the historical preference items of the current user are traversed, and the historical preference items are combined to form a historical preference set E of the user _p,0 (ii) a Then, with historical preference set E _p,0 Each entity node in the first layer diffusion preference set E is taken as a starting point, is connected to the corresponding associated entity of the entity through the knowledge graph, and is statistically formed _p,1 (ii) a Thereafter, the preference set E is diffused in the first layer _p,1 Each entity node in the second layer diffusion preference set E is taken as a starting point, is connected to the corresponding associated entity of the entity through the knowledge graph, and is statistically formed _p,2 . Namely, two adjacent diffusion preference sets, each entity of the diffusion preference set of the previous layer is connected to the corresponding entity of the diffusion preference set of the next layer through the knowledge graph. Finally, set of historical preferences E _p,0 First layer diffusion preference set E _p,1 And a second set of layer diffusion preferences E _p,2 Fusing to obtain the integral diffusion preference set E of the user _p . Wherein the bulk diffusion set E _p For the union of the diffusion preference sets for the respective layers, the size of each set may be set to be the same for simplicity of calculation. It is noted that the diffusion isThe number of layers can be adjusted according to actual conditions, and the larger the entity interval hop number is, the weaker the correlation is, so that in order to avoid the interference of the weak relationship with the recommendation result, only two-layer diffusion preference sets are calculated.

The mode of diffusing the historical preference data based on the knowledge graph not only considers the historical preference data, but also takes the user vector representation as the weighted sum of the entity vector representation in the user diffusion preference set, thereby effectively expanding the preference data of the user, avoiding the problem that new items can not be recommended, and being beneficial to improving the satisfaction degree of the user on the recommendation result.

And step S4: and inputting the user preference vector in the overall diffusion preference set into a preset RNN recommendation model for prediction to obtain a second recommendation neighbor set.

In the above step, the RNN recommendation model includes three layers of networks, an input layer, a hidden layer, and an output layer, where the input layer at the bottom layer is used to input the user preference vector in the whole diffusion preference set of the user. The middle hidden layer adopts a gate control cycle unit as an RNN unit, and the last output layer is used for outputting the probability that a current user is interested in a certain item, so that a second recommended neighbor set is obtained.

The hidden layer adopts a gate control circulation unit as an RNN unit, and the structure combines a forgetting gate and an input gate into a single updating gate, so that the structure is simpler, and the convergence rate of the model is higher. The update gate is used to determine the extent to which previous state information was passed into the current state, the reset gate is used to determine the extent to which previous state information is ignored, and the current state is a linear interpolation between the previous state and the candidate state. The RNN input sequence in the step is not a simple time sequence relation, but a hierarchical relation based on a user diffusion preference set, and items of an internal hierarchy are positioned behind the input sequence, so that deeper user preferences can be obtained, more auxiliary information is considered, and the accuracy of a recommendation result is improved.

The RNN recommendation model adopts an attention mechanism based on articles, different input parts are linearly combined according to the hierarchical relation of a user diffusion preference set to form preference characteristic representation of a user, so that the probability that the user is interested in a certain item is predicted, and the model is trained by minimizing cross entropy loss between a true value and a predicted value. Wherein the loss function is defined as follows:

wherein q is the predicted probability distribution, p is the true probability distribution, and N is the number of users.

Step S5: and fusing the first recommended neighbor set and the second recommended neighbor set to obtain a final recommended set, and pushing results in the recommended set to the user.

In the above steps, firstly, the recommendation results in the first recommendation neighbor set and the second recommendation neighbor set are sorted in a descending order according to the user score; then, extracting a plurality of previous recommendation results from the first recommendation neighbor set and the second recommendation neighbor set, and combining to form a push list; and then, screening and cleaning recommendation results in the push list to screen out repeatedly recommended items, so that a Top-N recommendation set is obtained, recommended related items are pushed to the user, and personalized recommendation of the radio and television technical knowledge is realized.

Referring to fig. 4, fig. 4 is a block diagram of an electronic device according to an embodiment of the present disclosure. The electronic device comprises a memory 1, a processor 2 and a communication interface 3, wherein the memory 1, the processor 2 and the communication interface 3 are electrically connected with each other directly or indirectly to realize the transmission or interaction of data. For example, the components may be electrically connected to each other via one or more communication buses or signal lines. The memory 1 can be used for storing software programs and modules, such as program instructions/modules corresponding to the broadcasting and television technology knowledge recommendation method based on user portrait and knowledge map provided by the embodiment of the application, and the processor 2 executes various functional applications and data processing by executing the software programs and modules stored in the memory 1. The communication interface 3 may be used for communication of signaling or data with other node devices.

The Memory 1 may be, but is not limited to, a Random Access Memory (RAM), a Read Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Read-Only Memory (EPROM), an electrically Erasable Read-Only Memory (EEPROM), and the like.

The processor 2 may be an integrated circuit chip having signal processing capabilities. The Processor 2 may be a general-purpose Processor including a Central Processing Unit (CPU), a Network Processor (NP), and the like; but also Digital Signal Processors (DSPs), application Specific Integrated Circuits (ASICs), field Programmable Gate Arrays (FPGAs) or other Programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components.

It will be appreciated that the configuration shown in fig. 4 is merely illustrative and that the electronic device may include more or fewer components than shown in fig. 4 or have a different configuration than shown in fig. 1. The components shown in fig. 4 may be implemented in hardware, software, or a combination thereof.

In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. The apparatus embodiments described above are merely illustrative, and for example, the flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

In addition, functional modules in the embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

The above-described functions, if implemented in the form of software functional modules and sold or used as a separate product, may be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the above-described method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made to the present application by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.

It will be evident to those skilled in the art that the present application is not limited to the details of the foregoing illustrative embodiments, and that the present application may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the application being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Claims

1. A broadcasting and TV technology knowledge recommendation method based on user portrait and knowledge map is characterized by comprising the following steps:

2. The broadcasting and television technology knowledge recommendation method based on the user portrait and the knowledge map as claimed in claim 1, wherein the step of obtaining the basic attribute information and the behavior feature information of the user, quantizing, fusing and reconstructing the information, and forming the user feature vector comprises:

3. The broadcasting and television technology knowledge recommendation method based on the user portrait and the knowledge graph as claimed in claim 2, wherein the step of calculating a similarity result between users by using a collaborative filtering algorithm based on the user feature vector, and performing collaborative recommendation according to the similarity result to obtain the first recommended neighbor set comprises:

according to the basic attribute vector and the user behavior vector of the user, respectively calculating attribute similarity and behavior similarity between the users by using a cosine similarity algorithm;

fusing the attribute similarity and the behavior similarity according to a fusion formula to obtain a similarity result between users so as to obtain similar users corresponding to the users through matching;

and acquiring a broadcasting and television content item table corresponding to the similar user and recommending the broadcasting and television content item table to the user to form a first recommended neighbor set.

4. The broadcasting and TV technology knowledge recommendation method based on user portrait and knowledge map as claimed in claim 3, wherein the fusion formula is:

5. The broadcasting and TV technology knowledge recommendation method based on user portrait and knowledge graph as claimed in claim 1, further comprising:

6. The broadcasting and television technology knowledge recommendation method based on the user portrait and the knowledge graph as claimed in claim 1, wherein the step of performing diffusion propagation in the pre-constructed knowledge graph of the broadcasting and television field according to historical preference data to obtain the overall diffusion preference set of the user comprises:

traversing historical preferences of the user, and combining to form a historical preference set of the user;

7. The radio and television technology knowledge recommendation method based on the user portrait and the knowledge graph as claimed in claim 1, wherein the RNN recommendation model adopts an article-based attention mechanism, different input parts are linearly combined according to a hierarchical relation of a user diffusion preference set to form a preference feature representation of the user so as to predict preference items of the user, and the model is trained by minimizing cross entropy loss between a true value and a predicted value.

8. The radio and television technology knowledge recommendation method based on the user portrait and the knowledge graph as claimed in claim 1, wherein the step of fusing the first recommended neighbor set and the second recommended neighbor set to obtain a final recommended set and pushing results in the recommended set to the user comprises:

extracting a plurality of previous recommendation results from the first recommendation adjacent set and the second recommendation adjacent set to form a push list;

9. An electronic device, comprising:

a memory for storing one or more programs;

a processor;

the one or more programs, when executed by the processor, implement the method of any of claims 1-8.

10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1-8.