CN115269998A - Information recommendation method and device, electronic equipment and storage medium - Google Patents
Information recommendation method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN115269998A CN115269998A CN202210998899.5A CN202210998899A CN115269998A CN 115269998 A CN115269998 A CN 115269998A CN 202210998899 A CN202210998899 A CN 202210998899A CN 115269998 A CN115269998 A CN 115269998A
- Authority
- CN
- China
- Prior art keywords
- information
- user
- probability
- candidate
- target
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application provides an information recommendation method and device, electronic equipment and a storage medium, and relates to the technical field of information processing. According to the method, the probability of recommending the candidate information to the user can be obtained by acquiring the candidate information and some information of the user, including historical reading information, behavior characteristics and target reading information, and analyzing and predicting according to the information by using a target recommendation model.
Description
Technical Field
The present application relates to the field of information processing technologies, and in particular, to an information recommendation method and apparatus, an electronic device, and a storage medium.
Background
With the rapid development of recommendation algorithms, recommendation systems have been widely applied to various application scenarios, for example, to recommend goods of interest to users in goods recommendation, and for example, to recommend news of interest to users in news recommendation.
However, the current recommendation method generally searches for some information similar to the information historically browsed by the user for recommendation, for example, the relevance between the information and the information is judged only according to the similarity of the information, so that it is difficult to find the potential interest habits of the user, which causes that the recommended information is difficult to meet the requirements of the user, and the accuracy of information recommendation cannot be ensured.
Disclosure of Invention
An embodiment of the present application aims to provide an information recommendation method, an information recommendation apparatus, an electronic device, and a storage medium, so as to solve a problem that information recommended by an existing recommendation method is inaccurate.
In a first aspect, an embodiment of the present application provides an information recommendation method, where the method includes:
acquiring candidate information and user information of a user, wherein the user information comprises historical reading information, behavior characteristics and target reading information of the user, and the target reading information refers to information which belongs to the same type as the candidate information in the historical reading information of the user;
and acquiring the probability of recommending the candidate information to the user according to the candidate information and the user information by using a target recommendation model.
In the implementation process, the probability of recommending the candidate information to the user can be obtained by acquiring the candidate information and some information of the user, including historical reading information, behavior characteristics and target reading information, and analyzing and predicting according to the information by using a target recommendation model.
Optionally, the obtaining, by using the target recommendation model, a probability of recommending the candidate information to the user according to the candidate information and the user information includes:
acquiring a first probability of recommending the candidate information to the user according to the candidate information and the historical reading information by using a target recommendation model;
acquiring a second probability of recommending the candidate information to the user according to the candidate information and the target reading information by using a target recommendation model;
obtaining a third probability of recommending the candidate information to the user according to the behavior characteristics by using a target recommendation model;
determining a final probability of recommending the candidate information to the user based on the first probability, the second probability, and the third probability using a target recommendation model.
In the implementation process, different information is respectively utilized to obtain corresponding recommendation probabilities, and the recommendation probabilities are finally integrated into the final recommendation probability, so that the potential behavior interests of the user can be mined from different angles, and more accurate recommendation is realized.
Optionally, the target recommendation model includes a language model and a user encoder, and the obtaining, by using the target recommendation model, a first probability of recommending the candidate information to the user according to the candidate information and the historical reading information includes:
respectively encoding the candidate information and the historical reading information by using the language model to obtain corresponding candidate information encoding vectors and historical reading information encoding vectors;
calculating the historical reading information coding vector and the position vector corresponding to the historical reading information by using the user encoder to obtain a first user vector;
and obtaining a first probability of recommending the candidate information to the user according to the first user vector and the candidate information coding vector.
In the implementation process, the language model and the user encoder can be adopted to effectively extract deep semantic information in the input information, and further more accurate recommendation probability can be predicted.
Optionally, the obtaining, by using the target recommendation model according to the candidate information and the target reading information, a second probability of recommending the candidate information to the user includes:
respectively encoding the candidate information and the target reading information by using the language model to obtain corresponding candidate information encoding vectors and target reading information encoding vectors;
calculating the target reading information coding vector and the position vector corresponding to the target reading information by using the user encoder to obtain a second user vector;
and obtaining a second probability of recommending the candidate information to the user according to the second user vector and the candidate information coding vector.
In the implementation process, the language model and the user encoder can be adopted to effectively extract deep semantic information in the input information, and further more accurate recommendation probability can be predicted.
Optionally, the language model is an electrora model, and since the model has smaller parameters, the calculation efficiency can be effectively improved, and a better information extraction effect is achieved.
Optionally, the target recommendation model further includes a prediction model, and the determining, by using the target recommendation model, a final probability of recommending the candidate information to the user based on the first probability, the second probability, and the third probability includes:
determining, with the predictive model, a final probability of recommending the candidate information to the user based on the first probability, the second probability, and the third probability.
In the implementation process, the three probabilities can be effectively integrated into the final recommendation probability by adopting the prediction model, so that more accurate probability prediction can be realized.
Optionally, the prediction model is an XGBoost model, which is a machine learning model, and can effectively learn the association between the three probabilities and the final probability, so as to realize more accurate probability prediction.
Optionally, the behavior feature includes at least one of a category number of the user reading information, a history information category reading probability, an information hot spot feature, an entity hot spot feature, and a user reading entity hot spot feature. By acquiring the behavior characteristics, the behavior interest of the user can be better extracted, and further more accurate recommendation is realized.
Optionally, the obtaining, by using a target recommendation model, a third probability that the candidate information is recommended to the user according to the behavior feature includes:
converting each characteristic contained in the behavior characteristic into a corresponding characteristic vector by using a target recommendation model;
splicing the feature vectors corresponding to the features by using the target recommendation model to obtain behavior feature vectors;
and inputting the behavior characteristic vector into a full connection layer by using the target recommendation model, and calculating by using a sigmoid function to obtain a third probability of recommending the candidate information to the user.
In the implementation process, the feature vectors corresponding to various features in the behavior features are spliced, then the spliced feature vectors are input into the full-connection layer and the sigmoid function is used for calculating to obtain the third probability, so that the behavior interest of the user can be better extracted, and more accurate recommendation is further realized.
Optionally, the candidate information is candidate news information, the historical reading information is historical reading news information, and the target reading information is target reading news information. The method and the device are applied to the field of news recommendation, and can realize accurate recommendation of news.
In a second aspect, an embodiment of the present application provides an information recommendation apparatus, where the apparatus includes:
the information acquisition module is used for acquiring candidate information and user information of a user, wherein the user information comprises historical reading information, behavior characteristics and target reading information of the user, and the target reading information refers to information which belongs to the same type as the candidate information in the historical reading information of the user;
and the information recommendation module is used for acquiring the probability of recommending the candidate information to the user according to the candidate information and the user information by using a target recommendation model.
In a third aspect, an embodiment of the present application provides an electronic device, including a processor and a memory, where the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps in the method as provided in the first aspect are executed.
In a fourth aspect, embodiments of the present application provide a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, performs the steps in the method as provided in the first aspect.
Additional features and advantages of the present application will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by the practice of the embodiments of the present application. The objectives and other advantages of the application may be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and that those skilled in the art can also obtain other related drawings based on the drawings without inventive efforts.
Fig. 1 is a flowchart of an information recommendation method according to an embodiment of the present application;
FIG. 2 is a schematic diagram of a general structure of a target recommendation model according to an embodiment of the present disclosure;
FIG. 3 is a detailed structural diagram of a target recommendation model according to an embodiment of the present disclosure;
fig. 4 is a block diagram of an information recommendation apparatus according to an embodiment of the present application;
fig. 5 is a schematic structural diagram of an electronic device for executing an information recommendation method according to an embodiment of the present application.
Detailed Description
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.
It should be noted that the terms "system" and "network" in the embodiments of the present invention may be used interchangeably. The "plurality" means two or more, and in view of this, the "plurality" may also be understood as "at least two" in the embodiments of the present invention. "and/or" describes the association relationship of the associated objects, meaning that there may be three relationships, e.g., a and/or B, which may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" generally indicates that the preceding and following related objects are in an "or" relationship, unless otherwise specified.
The embodiment of the application provides an information recommendation method, the method can obtain the probability of recommending candidate information to a user by acquiring the candidate information and some information of the user, including historical reading information, behavior characteristics and target reading information, and performing analysis prediction according to the information by using a target recommendation model.
Referring to fig. 1, fig. 1 is a flowchart of an information recommendation method according to an embodiment of the present application, where the method includes the following steps:
step S110: and acquiring candidate information and user information of the user, wherein the user information comprises historical reading information, behavior characteristics and target reading information of the user.
The execution main body of the information recommendation method in the application can be a background server or a front-end device, if the execution main body is the background server, the background server can send the obtained probability to the front-end device after obtaining the probability of recommending the candidate information to the user, the front-end device can determine whether to recommend the candidate information to the user according to the probability, or the background server can determine whether to recommend the candidate information to the user according to the probability after obtaining the probability, and then the judgment result can be sent to the front-end device, after the front-end device can obtain the judgment result, the front-end device can directly determine whether to recommend the candidate information to the user, and if so, the candidate information is displayed to the user. If the execution subject of the information recommendation method in the application is the front-end device, the steps can be executed by the front-end device, and after the probability is obtained, whether the candidate information needs to be recommended to the user can be determined according to the probability, and if so, the candidate information is displayed to the user.
It can be understood that the user referred to in the present application may refer to any user that needs to perform information recommendation, and for different users, if the same candidate information is obtained, the obtained user information is different, and the finally obtained recommendation probabilities may also be different, so that the method of the present application may implement accurate recommendation for different users.
The candidate information may refer to information to be recommended, for example, in some application scenarios, the candidate information is candidate news information, the historical reading information of the user is historical reading news information, and the target reading information is target reading news information. It can be understood that in other application scenarios, the candidate information may also be other information, such as candidate commodity information, candidate articles, candidate novels, candidate videos, and the like, and when information recommendation is involved in practical application, recommendation of corresponding information may be implemented by using the information recommendation method of the present application. For convenience of description, the following embodiments are illustrated in the application scenario of news recommendation.
The candidate information may be a set of information, for example, for news information, the candidate information may be a plurality of news, and for each news, the probability of whether to recommend the news to the user may be obtained according to the scheme provided in the present application.
The historical reading information of the user can be obtained from the historical browsing records of the user, for example, the historical browsing records of the user are crawled on a news website, the historical reading news information of the user and the behavior characteristics of the user can be extracted from the historical browsing records, and the behavior characteristics of the user refer to some characteristics representing the interests and hobbies of the user. And then acquiring some candidate news information, and extracting the read news information which belongs to the same type as the candidate news information by matching the historical read news information of the user with the candidate news information to be used as target read news information, namely the target read information in the application refers to the information which is used in the historical read information and belongs to the same type as the candidate information. The same type matching between the historical reading news information and the candidate news information can be performed through category matching, for example, the news information is entertainment news, or current news, or other ways to determine whether the news information and the candidate news information belong to the same category.
Step S120: and acquiring the probability of recommending the candidate information to the user according to the candidate information and the user information by using the target recommendation model.
In order to obtain a better recommendation effect, the target recommendation model adopted in the embodiment of the application can predict the probability of recommending the candidate information according to the obtained candidate information, the historical reading information, the behavior characteristics and the target reading information, and the target recommendation model can deeply extract the interest and hobby characteristics of the user from the candidate information, the historical reading information, the behavior characteristics and the target reading information, so that more accurate recommendation is realized.
In an application scenario of real-time recommendation, after the probability is obtained, when the probability is greater than or equal to a set probability (the set probability can be flexibly set according to actual requirements), the candidate information is determined to be recommended to the user, at this time, the candidate information can be output to the user, otherwise, if the probability is less than the set probability, the candidate information is determined not to be recommended to the user. For example, for news information, if not recommended, the news information may be hidden or sorted at a later position, and if recommended, the news information may be preferentially shown to the user. When there are multiple pieces of news information, for some pieces of news information to be recommended, the news information may be ranked according to the probability of recommending the news information, for example, the news information with the highest probability is ranked in the first place, and the news information with the lowest probability is ranked in the last place, and then the news information may be presented to the user according to the ranking order, so as to facilitate the user to read. Of course, in practical application, how to show or output the recommended candidate information to the user may be flexibly set according to actual requirements, and the embodiment of the present application is not particularly limited.
In the implementation process, the probability of recommending the candidate information to the user can be obtained by acquiring the candidate information and some information of the user, including historical reading information, behavior characteristics and target reading information, and analyzing and predicting according to the information by using a target recommendation model.
On the basis of the above embodiment, in the manner of obtaining the probability of recommending candidate information to the user, the target recommendation model may respectively obtain the probabilities by using various information, and then synthesize the probabilities, for example, a first probability of recommending candidate information to the user is obtained by using the target recommendation model according to the candidate information and the historical reading information, a second probability of recommending candidate information to the user is obtained by using the target recommendation model according to the candidate information and the target reading information, a third probability of recommending candidate information to the user is obtained by using the target recommendation model according to the behavior characteristics, and finally a final probability of recommending candidate information to the user is determined by using the target recommendation model based on the first probability, the second probability, and the third probability.
The historical reading information can represent the historical reading habits of the user, so that a first probability of recommending candidate information is predicted according to the historical reading information and the candidate information, the target reading information can represent the preference of the user for reading the same kind of information, a second probability of recommending candidate information can be predicted according to the target reading information and the candidate information, behavior characteristics of the user can represent some reading behaviors of the user, a third probability of recommending candidate information can be predicted according to the behavior characteristics, and finally the three probabilities can be integrated to obtain a final probability, for example, the three probabilities are weighted and averaged to obtain the final probability, wherein the weight can be obtained by training a target recommendation model, and can also be set manually in advance according to experience.
It can be understood that, here, the final probability may be determined according to the first probability, the second probability and the third probability, and may also be determined not only by a weighted average, but also by other manners, such as a direct average, or a manner in which the maximum probability is taken as the final probability, and in practical applications, what manner is specifically adopted may be verified in advance through continuous experiments, and finally, a manner with the best recommendation effect may be adopted.
In the implementation process, different information is respectively utilized to obtain corresponding recommendation probabilities, and the recommendation probabilities are finally integrated into the final recommendation probability, so that the potential behavior interests of the user can be mined from different angles, and more accurate recommendation is realized.
On the basis of the foregoing embodiment, as shown in fig. 2, an exemplary embodiment of the present invention provides a structural diagram of a target recommendation model, where the target recommendation model includes a language model and a user encoder, and in the manner of obtaining the first probability, the language model may be used to encode the candidate information and the historical reading information respectively to obtain a corresponding candidate information encoding vector and a historical reading information encoding vector, and the user encoder is used to calculate the historical reading information encoding vector and a position vector corresponding to the historical reading information to obtain a first user vector, and then a first probability of recommending the candidate information to the user is obtained according to the first user vector and the candidate information encoding vector.
If there are multiple pieces of history reading information, each piece of history reading information can be converted into a corresponding coding vector by the language model (of course, each piece of history reading information can be input into one language model, that is, there are multiple language models, so that the calculation efficiency can be improved), and then the coding vectors are input into the user encoder. The user encoder is used for calculating a historical reading information encoding vector and a corresponding position vector thereof, wherein the position vector refers to a vector formed by the positions of all words in the historical reading information so as to be used for identifying semantic information in the historical reading information. The vector calculation may be understood as that the user encoder converts a plurality of input vectors into an intermediate representation vector through a certain calculation, that is, converts the historical reading information encoding vector and the position vector into an intermediate first user vector after calculation.
Here, the manner of obtaining the first probability according to the first user vector and the candidate information coding vector may be to perform cosine similarity calculation on the first user vector and the candidate information coding vector, where the obtained similarity may be used as the first probability, and of course, the euclidean distance calculation may also be performed on the first user vector and the candidate information coding vector, and the obtained distance may be used as the first probability. It is understood that the first probability may also be obtained by performing similarity calculation between the first user vector and the candidate information encoding vector in other manners, and in practical applications, the similarity calculation manner may be flexibly selected to obtain the first probability.
Similarly, in the above manner of obtaining the second probability, the language model may be used to encode the candidate information and the target reading information respectively to obtain a corresponding candidate information encoding vector and a corresponding target reading information encoding vector, then the user encoder is used to calculate the target reading information encoding vector and the position vector corresponding to the target reading information to obtain a second user vector, and a second probability of recommending the candidate information to the user is obtained according to the second user vector and the candidate information encoding vector.
The method for encoding the candidate information and the target reading information by using the language model is similar to the usage of the language model in the above embodiment, and the method for calculating the target reading information encoding vector and the corresponding position vector by using the user encoder is also similar to the usage of the user encoder in the above embodiment, which will not be described herein. The manner of obtaining the second probability according to the second user vector and the candidate information coding vector may also be a manner of performing cosine similarity calculation, euclidean distance calculation, and the like on the second user vector and the candidate information coding vector, and the obtained cosine similarity or euclidean distance may be used as the second probability.
In the implementation process, the language model and the user encoder can be adopted to effectively extract deep semantic information in the input information, and further more accurate recommendation probability can be predicted.
On the basis of the above-described embodiment, the language model may be BERT (Bidirectional Encoder Representation from transforms, encoder for bi-directional Transformers) model that is capable of extracting rich semantic information representations in the input text.
In order to improve recommendation efficiency in the embodiment of the application, the language model may adopt an elctra (efficient Learning Encoder for Accurately classifying the label replacement) model, and because the elctra model is compared with the BERT model, the elctra model has the advantages of smaller parameters, higher calculation efficiency and better effect, and therefore, the elctra model can be used for realizing coding conversion of input information more quickly.
It can be understood that the language model in the embodiment of the present application may also adopt other language models, and in practical application, the corresponding language model may be flexibly selected according to practical requirements.
The user encoder in the embodiment of the present application may adopt a conventional encoder, for example, the user encoder includes a stack of a plurality of layers, each layer includes two sublayers, one of the sublayers is a network layer of a multi-head attention mechanism (a plurality of attention modules are used in parallel in the multi-head attention mechanism, which can enrich the diversity of attention and increase the expression capability of the model), the other sublayer is a fully-connected feedforward neural network layer, and the two sublayers are connected through a residual error network and a normalization layer.
Wherein, the input of the multi-head attention mechanism network layer is connected with the output of the language model, and the output of the feedforward neural network layer is connected with other network structures of the target recommendation model, such as the output of the feedforward neural network layer is connected with a probability calculation layer (such as the first probability calculation or the second probability calculation in fig. 2).
The number of the stacks of the plurality of layers may be flexibly set according to actual requirements, for example, the number of the stacks of 6 layers is not particularly limited in this embodiment of the present application.
On the basis of the above embodiment, the behavior characteristics of the user include at least one of the category number of the user reading information, the history information category reading probability, the information hotspot characteristics, the entity hotspot characteristics, and the user reading entity hotspot characteristics.
The following describes a process of obtaining a third probability of recommending candidate information to the user according to the behavior feature.
Taking a news recommendation scene as an example, first, statistics can be performed on N types of news read by the user in the user log, and the total is marked as C = (C) 1 ,c 2 ,...,c N )。
History news category reading probability (history information category reading probability): probability p i For a given c i ，c i And e, belonging to a news in the C class, calculating the probability of the news in all the read news of the user, and recording the probability as P.
News hotspot characteristic (information hotspot characteristic): for a given c i ，c i E to a news in C class, and counting the reading quantity R of the news a And the historical reading total amount of the news
The news hotspotIs characterized by passing through
And (6) calculating.
Entity hotspot characteristics: for a given c i ，c i E, news in class C, and counting the entity reading amount R for given candidate news e Historical reading total amount with the entity
Entity hot spot feature passing of this type of news
And (6) calculating.
Reading entity hot spot characteristics by a user: counting the occurrence frequency R of a news entity in the historical news of a user t And total number of news read R per user h User reading entity news hotspot feature is through F t ＝R t /R h And (6) calculating.
During the analysis, it was found that the reading interests of the user changed over time, but followed the general trend of news events. In order to make the recommendation system more time-efficient, two-dimensional features are added, that is, on the basis of the total number of user behavior features and the reading probability in the class (that is, the reading probability of the historical information class), for example, the latest 20 user behavior features are counted and are respectively marked as C 'and P'.
Total number of last 20 user behavior features: in the statistical user log, the total N 'news recorded as C' = (C ') is recorded according to the latest 20 reads of the user' 1 ,c' 2 ,...,c' N )。
Probability of reading of the last 20 historical news categories: probability p' i Indicates for a given c i ，c i And e, belonging to a news in the C class, calculating the probability of the news in the latest 20 read news of the user, and recording the probability as P'.
Then the characteristics can be converted into corresponding characteristic vectors through a behavior encoder, the characteristic vectors corresponding to the characteristics are spliced to obtain a behavior characteristic vector,i.e. Concat (C, P, C ', P', F) a ,F e ,F t ) (it can be understood that if there are only one or more of the behavior features of the user, the spliced vectors need to be adjusted accordingly), and finally, the behavior feature vectors obtained by splicing are input into the full connection layer and are calculated through the sigmoid function, so that a third probability that candidate news is recommended to the user can be obtained. By acquiring the behavior characteristics, the behavior interest of the user can be better extracted, and further more accurate recommendation is realized.
On the basis of the foregoing embodiment, as shown in fig. 2, the target recommendation model may further include a prediction model, and the prediction model may be configured to predict a final probability of recommending the candidate information to the user, that is, determine the final probability of recommending the candidate information to the user based on the first probability, the second probability, and the third probability by using the prediction model.
The prediction model may be a weighted average model mentioned in the above embodiments, or may be other probability calculation models. In order to achieve a better prediction effect, the prediction model may also be a machine learning model, such as an XGBoost (eXtreme Gradient Boosting) model, and the first probability, the second probability and the third probability may be better fused through the XGBoost model to obtain a more accurate recommendation probability.
The target recommendation model in the application may be formed by other network layers such as an electrora model, a user encoder, an XGBoost model, and the like, and a detailed structure diagram may be as shown in fig. 3.
In addition, before information recommendation, a large number of training samples can be used for training the target recommendation model, in the training process, the target recommendation model can be trained integrally, in order to improve the training efficiency, three models in the target recommendation model can be trained respectively, namely an electrora model, a user encoder and an XGBoost model are trained respectively, for example, the electrora model is trained firstly, after the training is completed, network parameters of the electrora model are solidified, then the user encoder is trained, after the training is completed, the network parameters of the electrora model and the user encoder are solidified, then the XGBoost model is trained, and after the model training is completed, the trained target recommendation model can be obtained. Thus, by step-by-step training, the influence of adjusting the parameters of one model on the results of other models can be avoided in the training process, thus, the training efficiency can be greatly improved.
Referring to fig. 4, fig. 4 is a block diagram of an information recommendation apparatus 200 according to an embodiment of the present disclosure, where the apparatus 200 may be a module, a program segment, or a code on an electronic device (e.g., a backend server or a front-end device). It should be understood that the apparatus 200 corresponds to the above-mentioned embodiment of the method of fig. 1, and can perform various steps related to the embodiment of the method of fig. 1, and the specific functions of the apparatus 200 can be referred to the above description, and the detailed description is appropriately omitted here to avoid redundancy.
Optionally, the apparatus 200 comprises:
the information obtaining module 210 is configured to obtain candidate information and user information of a user, where the user information includes historical reading information, behavior characteristics, and target reading information of the user, and the target reading information is information in the historical reading information of the user, which is of the same type as the candidate information;
and the information recommendation module 220 is configured to acquire, by using a target recommendation model, a probability of recommending the candidate information to the user according to the candidate information and the user information.
Optionally, the information recommending module 220 is configured to obtain, by using a target recommendation model, a first probability of recommending the candidate information to the user according to the candidate information and the historical reading information; acquiring a second probability of recommending the candidate information to the user according to the candidate information and the target reading information by using a target recommendation model; obtaining a third probability of recommending the candidate information to the user according to the behavior characteristics by using a target recommendation model; determining a final probability of recommending the candidate information to the user based on the first probability, the second probability, and the third probability using a target recommendation model.
Optionally, the target recommendation model includes a language model and a user encoder, and the information recommendation module 220 is configured to encode the candidate information and the historical reading information by using the language model respectively to obtain corresponding candidate information encoding vectors and historical reading information encoding vectors; calculating the historical reading information coding vector and the position vector corresponding to the historical reading information by using the user encoder to obtain a first user vector; and obtaining a first probability of recommending the candidate information to the user according to the first user vector and the candidate information coding vector.
Optionally, the target recommendation model includes a language model and a user encoder, and the information recommendation module 220 is configured to encode the candidate information and the target reading information respectively by using the language model to obtain a corresponding candidate information encoding vector and a corresponding target reading information encoding vector; calculating the target reading information coding vector and the position vector corresponding to the target reading information by using the user encoder to obtain a second user vector; and obtaining a second probability of recommending the candidate information to the user according to the second user vector and the candidate information coding vector.
Optionally, the language model is an ectra model.
Optionally, the target recommendation model further includes a prediction model, and the information recommendation module 220 is configured to determine a final probability of recommending the candidate information to the user based on the first probability, the second probability, and the third probability by using the prediction model.
Optionally, the prediction model is an XGBoost model.
Optionally, the behavior feature includes at least one of a category number of the user reading information, a history information category reading probability, an information hot spot feature, an entity hot spot feature, and a user reading entity hot spot feature.
Optionally, the information recommendation module 220 is configured to convert each feature included in the behavior feature into a corresponding feature vector by using a target recommendation model; splicing the feature vectors corresponding to the features by using the target recommendation model to obtain behavior feature vectors; and inputting the behavior characteristic vector into a full connection layer by using the target recommendation model, and calculating by using a sigmoid function to obtain a third probability of recommending the candidate information to the user.
Optionally, the candidate information is candidate news information, the historical reading information is historical reading news information, and the target reading information is target reading news information.
It should be noted that, for the convenience and simplicity of description, the specific working process of the above-described device may refer to the corresponding process in the foregoing method embodiment, and the description is not repeated here.
Referring to fig. 5, fig. 5 is a schematic structural diagram of an electronic device for executing an information recommendation method according to an embodiment of the present application, where the electronic device may include: at least one processor 310, such as a CPU, at least one communication interface 320, at least one memory 330, and at least one communication bus 340. Wherein the communication bus 340 is used for realizing direct connection communication of these components. The communication interface 320 of the device in the embodiment of the present application is used for performing signaling or data communication with other node devices. The memory 330 may be a high-speed RAM memory or a non-volatile memory (e.g., at least one disk memory). The memory 330 may optionally also be at least one storage device located remotely from the aforementioned processor. The memory 330 stores computer readable instructions, which when executed by the processor 310, cause the electronic device to perform the method processes described above with reference to fig. 1.
It will be appreciated that the configuration shown in fig. 5 is merely illustrative and that the electronic device may include more or fewer components than shown in fig. 5 or may have a different configuration than shown in fig. 5. The components shown in fig. 5 may be implemented in hardware, software, or a combination thereof.
Embodiments of the present application provide a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, performs the method processes performed by an electronic device in the method embodiment shown in fig. 1.
The present embodiments disclose a computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions which, when executed by a computer, enable the computer to perform the methods provided by the above-described method embodiments, for example, comprising:
acquiring candidate information and user information of a user, wherein the user information comprises historical reading information, behavior characteristics and target reading information of the user, and the target reading information refers to information which belongs to the same type as the candidate information in the historical reading information of the user;
and acquiring the probability of recommending the candidate information to the user according to the candidate information and the user information by using a target recommendation model.
In summary, the embodiment of the present application provides an information recommendation method, an information recommendation device, an electronic device, and a storage medium, in which the method obtains candidate information and some information of a user, including historical reading information, behavior characteristics, and target reading information, and performs analysis and prediction according to the information by using a target recommendation model, so as to obtain a probability of recommending the candidate information to the user.
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 above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units is only one logical division, and there may be other divisions when actually implemented, and for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed coupling or direct coupling or communication connection between each other may be through some communication interfaces, indirect coupling or communication connection between devices or units, and may be in an electrical, mechanical or other form.
In addition, units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
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 alone, or two or more modules may be integrated to form an independent part.
In this document, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions.
The above description is only an example of the present application and is not intended to limit the scope of the present application, and various modifications and changes may be made 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.
Claims (13)
1. An information recommendation method, characterized in that the method comprises:
acquiring candidate information and user information of a user, wherein the user information comprises historical reading information, behavior characteristics and target reading information of the user, and the target reading information refers to information which belongs to the same type as the candidate information in the historical reading information of the user;
and acquiring the probability of recommending the candidate information to the user according to the candidate information and the user information by using a target recommendation model.
2. The method of claim 1, wherein the obtaining, by using the target recommendation model, the probability of recommending the candidate information to the user according to the candidate information and the user information comprises:
acquiring a first probability of recommending the candidate information to the user according to the candidate information and the historical reading information by using a target recommendation model;
acquiring a second probability of recommending the candidate information to the user according to the candidate information and the target reading information by using a target recommendation model;
obtaining a third probability of recommending the candidate information to the user according to the behavior characteristics by using a target recommendation model;
determining a final probability of recommending the candidate information to the user based on the first probability, the second probability, and the third probability using a target recommendation model.
3. The method of claim 2, wherein the target recommendation model comprises a language model and a user encoder, and the obtaining, by the target recommendation model, a first probability of recommending the candidate information to the user according to the candidate information and the historical reading information comprises:
coding the candidate information and the historical reading information by using the language model respectively to obtain corresponding candidate information coding vectors and historical reading information coding vectors;
calculating the historical reading information coding vector and the position vector corresponding to the historical reading information by using the user encoder to obtain a first user vector;
and obtaining a first probability of recommending the candidate information to the user according to the first user vector and the candidate information coding vector.
4. The method of claim 2, wherein the target recommendation model comprises a language model and a user encoder, and the obtaining, by the target recommendation model, a second probability of recommending the candidate information to the user according to the candidate information and the target reading information comprises:
respectively encoding the candidate information and the target reading information by using the language model to obtain corresponding candidate information encoding vectors and target reading information encoding vectors;
calculating the target reading information coding vector and the position vector corresponding to the target reading information by using the user encoder to obtain a second user vector;
and obtaining a second probability of recommending the candidate information to the user according to the second user vector and the candidate information coding vector.
5. The method of claim 3 or 4, wherein the language model is an Electrora model.
6. The method of claim 3 or 4, wherein the target recommendation model further comprises a prediction model, and wherein determining a final probability of recommending the candidate information to the user based on the first probability, the second probability, and the third probability using the target recommendation model comprises:
determining, with the predictive model, a final probability of recommending the candidate information to the user based on the first probability, the second probability, and the third probability.
7. The method of claim 6, wherein the predictive model is an XGboost model.
8. The method of claim 2, wherein the behavior characteristics comprise at least one of a category number of the user reading information, a history information category reading probability, an information hot spot characteristic, an entity hot spot characteristic, and a user reading entity hot spot characteristic.
9. The method of claim 8, wherein obtaining a third probability that the candidate information is recommended to the user according to the behavior feature by using a target recommendation model comprises:
converting each characteristic contained in the behavior characteristic into a corresponding characteristic vector by using a target recommendation model;
splicing the feature vectors corresponding to the features by using the target recommendation model to obtain behavior feature vectors;
and inputting the behavior characteristic vector into a full connection layer by using the target recommendation model, and calculating by using a sigmoid function to obtain a third probability of recommending the candidate information to the user.
10. The method of claim 1, wherein the candidate information is candidate news information, the historical reading information is historical reading news information, and the target reading information is target reading news information.
11. An information recommendation apparatus, characterized in that the apparatus comprises:
the information acquisition module is used for acquiring candidate information and user information of a user, wherein the user information comprises historical reading information, behavior characteristics and target reading information of the user, and the target reading information refers to information which belongs to the same type as the candidate information in the historical reading information of the user;
and the information recommendation module is used for acquiring the probability of recommending the candidate information to the user according to the candidate information and the user information by using a target recommendation model.
12. An electronic device comprising a processor and a memory, the memory storing computer readable instructions that, when executed by the processor, perform the method of any of claims 1-10.
13. 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 1-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210998899.5A CN115269998A (en) | 2022-08-19 | 2022-08-19 | Information recommendation method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210998899.5A CN115269998A (en) | 2022-08-19 | 2022-08-19 | Information recommendation method and device, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115269998A true CN115269998A (en) | 2022-11-01 |
Family
ID=83752164
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210998899.5A Pending CN115269998A (en) | 2022-08-19 | 2022-08-19 | Information recommendation method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115269998A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117150145A (en) * | 2023-10-31 | 2023-12-01 | 成都企软数字科技有限公司 | Personalized news recommendation method and system based on large language model |
-
2022
- 2022-08-19 CN CN202210998899.5A patent/CN115269998A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117150145A (en) * | 2023-10-31 | 2023-12-01 | 成都企软数字科技有限公司 | Personalized news recommendation method and system based on large language model |
| CN117150145B (en) * | 2023-10-31 | 2024-01-02 | 成都企软数字科技有限公司 | Personalized news recommendation method and system based on large language model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110309427B (en) | Object recommendation method and device and storage medium | |
| CN111797321B (en) | Personalized knowledge recommendation method and system for different scenes | |
| CN110717099B (en) | Method and terminal for recommending film | |
| CN112364204B (en) | Video searching method, device, computer equipment and storage medium | |
| CN110019794B (en) | Text resource classification method and device, storage medium and electronic device | |
| CN109948735B (en) | Multi-label classification method, system, device and storage medium | |
| CN110390052B (en) | Search recommendation method, training method, device and equipment of CTR (China train redundancy report) estimation model | |
| CN110083774B (en) | Method and device for determining application recommendation list, computer equipment and storage medium | |
| CN117197722B (en) | User perception and analysis system based on mobile internet video | |
| CN115269998A (en) | Information recommendation method and device, electronic equipment and storage medium | |
| CN113987161A (en) | Text sorting method and device | |
| CN113032676A (en) | Recommendation method and system based on micro-feedback | |
| CN110348581B (en) | User feature optimizing method, device, medium and electronic equipment in user feature group | |
| CN112330442A (en) | Modeling method and device based on ultra-long behavior sequence, terminal and storage medium | |
| CN115618079A (en) | Session recommendation method, device, electronic equipment and storage medium | |
| CN115203206A (en) | Data content searching method and device, computer equipment and readable storage medium | |
| CN114528908A (en) | Network request data classification model training method, classification method and storage medium | |
| CN114780757A (en) | Short media label extraction method and device, computer equipment and storage medium | |
| CN111611981A (en) | Information identification method and device and information identification neural network training method and device | |
| CN117786234B (en) | Multimode resource recommendation method based on two-stage comparison learning | |
| CN116628179B (en) | User operation data visualization and man-machine interaction recommendation method | |
| CN112000821B (en) | Multimedia information pushing method, device, server and storage medium | |
| Che et al. | A feature and deep learning model recommendation system for mobile application | |
| CN117391713A (en) | Information pushing method and device, electronic equipment and storage medium | |
| CN116720517A (en) | Search word component recognition model construction method and search word component recognition method |
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 |