CN113672804A - Recommendation information generation method, system, computer device and storage medium - Google Patents

Recommendation information generation method, system, computer device and storage medium Download PDF

Info

Publication number
CN113672804A
CN113672804A CN202110882147.8A CN202110882147A CN113672804A CN 113672804 A CN113672804 A CN 113672804A CN 202110882147 A CN202110882147 A CN 202110882147A CN 113672804 A CN113672804 A CN 113672804A
Authority
CN
China
Prior art keywords
user
advertisement
information
recommendation information
information generation
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
Application number
CN202110882147.8A
Other languages
Chinese (zh)
Inventor
张九龙
李锋
杨洋
张琛
万化
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Pudong Development Bank Co Ltd
Original Assignee
Shanghai Pudong Development Bank Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shanghai Pudong Development Bank Co Ltd filed Critical Shanghai Pudong Development Bank Co Ltd
Priority to CN202110882147.8A priority Critical patent/CN113672804A/en
Publication of CN113672804A publication Critical patent/CN113672804A/en
Pending legal-status Critical Current

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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0255—Targeted advertisements based on user history
    • G06Q30/0256—User search
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0269—Targeted advertisements based on user profile or attribute
    • G06Q30/0271—Personalized advertisement

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Accounting & Taxation (AREA)
  • Databases & Information Systems (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • General Engineering & Computer Science (AREA)
  • Game Theory and Decision Science (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Probability & Statistics with Applications (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Information Transfer Between Computers (AREA)

Abstract

The invention relates to the technical field of conversation systems, and discloses a recommendation information generation method, a system, computer equipment and a storage medium, wherein the recommendation information generation method comprises the steps of obtaining an advertisement identification mark of a target advertisement, determining a cluster type of the target advertisement according to the advertisement identification mark, and corresponding different cluster types to different pre-constructed recommendation information generation models; acquiring a user identification of a query user, and determining user interest information of the query user according to the user identification; generating user interest information in advance according to user behavior data of a user; inputting the user interest information and the query information of the query user into a recommendation information generation model corresponding to the cluster type to acquire the recommendation information of the target advertisement. By clustering advertisements in advance, a recommendation information generation model is individually constructed for each cluster type. The generation modes of the recommendation information of the advertisements in all cluster types after clustering are more similar, and the situation that the recommendation information generated by the recommendation information generation model is irrelevant to input can be reduced.

Description

Recommendation information generation method, system, computer device and storage medium
Technical Field
The present invention relates to the field of dialog system technologies, and in particular, to a recommendation information generation method, a recommendation information generation system, a computer device, and a storage medium.
Background
At present, recommendation information for searching advertisements mainly comprises manual operation, comment data extraction and content generation. However, recommendation information generated by means of manual operation and extraction is relatively pointless, while related schemes based on content generation have problems that easily result in generation of wrong/irrelevant recommendations, and the candidate set of advertisement scenes is relatively few.
Disclosure of Invention
Based on this, it is necessary to provide a recommendation information generation method, system, computer device, and storage medium for solving the problem that a correlation scheme based on content generation easily causes generation of erroneous/irrelevant recommenders.
A recommendation information generation method comprises the steps of obtaining an advertisement identification mark of a target advertisement, determining a cluster type of the target advertisement according to the advertisement identification mark, wherein the cluster type is obtained by clustering the advertisement identification mark according to question and answer data of the advertisement in advance, and different cluster types correspond to different recommendation information generation models which are constructed in advance; acquiring a user identification of a query user, and determining user interest information of the query user according to the user identification; the user interest information is generated in advance according to user behavior data of a user; and inputting the user interest information and the query information of the query user into a recommendation information generation model corresponding to the cluster type to acquire the recommendation information of the target advertisement.
In one embodiment, before clustering the advertisement identification marks according to question and answer data of the advertisements to obtain cluster types, the method further comprises formatting the question and answer data of the advertisements to obtain the question and answer data with a preset question and answer data format; and carrying out format arrangement on the user behavior data of the user to obtain the user behavior data with a preset behavior data format.
In one embodiment, the question-answer data includes question data and answer data, and the clustering of the advertisement identification marks in advance according to the question-answer data of the advertisements includes aggregating the advertisement identification marks corresponding to the answer data meeting preset similar conditions into a cluster to obtain different cluster types.
In one embodiment, the aggregating the advertisement identification identifiers corresponding to the answer data meeting the preset similar conditions into a cluster, and acquiring different cluster types includes converting the answer data into an answer data vector of a fixed dimension by using a paragraph vector method; and clustering the answer data vectors by adopting a K-means clustering algorithm, and clustering the advertisements into clusters with a preset number.
In one embodiment, before clustering the answer data vectors by using a K-means clustering algorithm and clustering the advertisements into a preset number of clusters, the method further includes setting an initial value of the preset number; after the advertisements are clustered into clusters with a preset number by adopting a K-means clustering algorithm, the method also comprises the step of adjusting the value of the preset number according to the clustering effect.
In one embodiment, the user behavior data includes a user identification and historical query information, and generating user interest information in advance according to the user behavior data of the user includes using the user behavior data as training data of a word vector model to generate user interest information, wherein one user corresponds to one user interest information.
In one embodiment, the user behavior data includes a user identification mark, an advertisement identification mark, historical query information and historical advertisement recommendation information, the different cluster types correspond to different pre-constructed recommendation information generation models, and the different cluster types include an advertisement identification mark corresponding to a target cluster type, user behavior data corresponding to the target cluster type is obtained according to the advertisement identification mark, and user interest information corresponding to the user identification mark is obtained according to the user identification mark in the user behavior data; and training the information generation model according to the user interest information corresponding to the user identification and the historical query information in the user behavior data corresponding to the target cluster type to obtain the recommendation information generation model corresponding to the target cluster type.
In one embodiment, the training of the information generation model according to the user interest information corresponding to the user identification and the historical query information in the user behavior data corresponding to the target cluster type to obtain the recommendation information generation model corresponding to the target cluster type includes taking the historical query information in the user behavior data corresponding to the target cluster type as the input of an encoder in the information generation model to obtain encoding output information; calculating the similarity information between the user interest information corresponding to the user identification and the coded output information; splicing the similarity information, the coding output information and the user interest information to obtain spliced information; taking the splicing information as the input of a decoder in the information generation model to acquire training recommendation information; and comparing the training recommendation information with historical advertisement recommendation information, and optimizing the information generation model according to the comparison result to obtain a recommendation information generation model corresponding to the target cluster type.
A recommendation information generation system comprises an advertisement cluster identification module, a recommendation information generation module and a recommendation information generation module, wherein the advertisement cluster identification module is used for acquiring an advertisement identification mark of a target advertisement, and determining a cluster type of the target advertisement according to the advertisement identification mark, the cluster type is obtained by clustering the advertisement identification mark according to question and answer data of the advertisement in advance, and different cluster types correspond to different recommendation information generation models which are constructed in advance; the user interest identification module is used for acquiring a user identification mark of a query user and determining user interest information of the query user according to the user identification mark; the user interest information is generated in advance according to user behavior data of a user; and the recommendation information generation module is used for inputting the user interest information and the query information of the query user into a recommendation information generation model corresponding to the cluster type so as to obtain the recommendation information of the target advertisement.
A computer device comprising a memory storing a computer program and a processor implementing the steps of the recommendation information generation method of any of the above embodiments when the processor executes the computer program.
A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, implements the steps of the recommendation information generation method according to any one of the above-mentioned embodiments.
According to the recommendation information generation method, different cluster types are obtained by clustering according to the question and answer data of the advertisements in advance, and a recommendation information generation model is independently constructed for each cluster type. The generation modes of the recommendation information of the advertisements in all cluster types after clustering are more similar, so that the relevance between the recommendation information generated by the recommendation information generation model and the query information input by the user is stronger, and the condition that the recommendation information generated by the recommendation information generation model is irrelevant to the input can be reduced. After the query user inputs query information and determines the advertisements needing to be pushed to the query user, the cluster type of the target advertisements can be determined according to the advertisement identification marks of the target advertisements, and therefore recommendation information of the target advertisements is generated by using a recommendation information generation model corresponding to the cluster type. Meanwhile, the recommendation information generated by the recommendation information generation model is personalized according to the user interest information corresponding to the query user and is more in line with the preference of the user, so that the recommendation effect of the recommendation information on the target advertisement is optimized.
Drawings
In order to more clearly illustrate the embodiments of the present specification or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the specification, and other drawings can be obtained by those skilled in the art without inventive labor.
Fig. 1 is a schematic flowchart of a method of generating recommendation information according to an embodiment of the present disclosure;
FIG. 2 is a schematic flow chart illustrating a method for data formatting according to an embodiment of the present disclosure;
FIG. 3 is a flowchart illustrating a method for aggregating the advertisement identifications with similar response data according to an embodiment of the present disclosure;
FIG. 4 is a flowchart illustrating a method for constructing different recommendation generation models, according to an embodiment of the present disclosure;
fig. 5 is a flowchart illustrating a method for obtaining a recommendation information generation model corresponding to a target cluster type according to an embodiment of the disclosure;
FIG. 6 is a schematic structural diagram of an information generating method system according to an embodiment of the present disclosure;
fig. 7 is a block diagram of a recommendation information generation apparatus or system according to an embodiment of the present disclosure.
Detailed Description
To facilitate an understanding of the invention, the invention will now be described more fully with reference to the accompanying drawings. Preferred embodiments of the present invention are shown in the drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
At present, schemes for pushing a text for an advertisement according to search contents of a user mainly include a manual operation method, a comment data extraction method and a content generation method. The manual operation method is that an operator or a specialist writes a proper text content for the advertisement of each merchant. The comment data extraction method is used for extracting part of comment content written by a user to a merchant to serve as recommendation information. The content generation method is to train a generation model by using NLP (Natural Language Processing) technology to automatically generate an appropriate text by the model.
However, the manual operation and the generation of the extracted comment data cannot answer the search content input by the user in a targeted manner. Most of the technical schemes based on content generation only adopt one model, which easily results in generation of wrong/irrelevant recommenders. Therefore, there is a need for a recommendation information generation method that can generate recommendation information more relevant to user input information.
Fig. 1 is a schematic flowchart of a method for generating recommendation information according to an embodiment of the present disclosure, where in an embodiment, the method for generating recommendation information includes the following steps S100 to S300.
Step S100: the method comprises the steps of obtaining advertisement identification marks of target advertisements, determining cluster types to which the target advertisements belong according to the advertisement identification marks, wherein the cluster types are obtained by clustering the advertisement identification marks according to question and answer data of the advertisements in advance, and different cluster types correspond to different pre-constructed recommendation information generation models.
Each advertisement corresponds to an advertisement identification mark. In the data processing completed in advance, clustering processing can be performed according to the question and answer data of each advertisement, and advertisements with certain similarity are clustered into a cluster type, so that a plurality of different cluster types are obtained. A mapping relation of < advertisement identification, cluster type > may be constructed from the clustering result. In addition, in the data processing in advance, a recommendation information generation model corresponding to each cluster type is also constructed. After the target advertisement needing to be displayed to the user is determined, the advertisement identification mark of the target advertisement is obtained, so that the cluster type corresponding to the target advertisement can be determined according to the mapping relation and the advertisement identification mark of the target advertisement. The cluster type corresponding to the target advertisement is determined, that is, the recommendation information generation process for the target advertisement can be completed by using the recommendation information generation model corresponding to the cluster type.
Step S200: acquiring a user identification of a query user, and determining user interest information of the query user according to the user identification; the user interest information is generated in advance according to user behavior data of the user.
Each user also corresponds to a user identification. In the data processing completed in advance, the user behavior data of the user in the history data may be processed to acquire the user interest information of the user. Each user corresponds to corresponding user interest information, and the user interest information can express the preference of the user, so that more personalized recommendation information can be generated by taking the user preference as a reference. According to the user identification mark of the query user who inputs the query information at present, the user interest information corresponding to the user identification mark can be determined.
Step S300: inputting the user interest information and the query information of the query user into a recommendation information generation model corresponding to the cluster type to acquire the recommendation information of the target advertisement.
And inputting the query information input by the query user and the user interest information corresponding to the query user into the recommendation information generation model corresponding to the cluster type. Because the question-answer mode of each advertisement which has a mapping relation with the cluster type after clustering is more similar, the situation that the generated recommendation information is irrelevant to the query information can be reduced by using the recommendation information generation model corresponding to the cluster type. Meanwhile, in the process of information analysis and generation, the recommendation information generation model introduces user interest information corresponding to the query user, so that the recommendation information generated by the recommendation information generation model is more personalized and better accords with the preference of the query user, and the click rate of the query user on the target advertisement can be improved.
According to the recommendation information generation method, the advertisements are clustered in advance, and a recommendation information generation model is independently constructed for each cluster type. After the query user inputs query information and determines the advertisements needing to be pushed to the query user, the cluster type of the target advertisements can be determined according to the advertisement identification marks of the target advertisements, and therefore recommendation information of the target advertisements is generated by using a recommendation information generation model corresponding to the cluster type. When the recommendation information generation model corresponding to the cluster type is used for generating the recommendation information of the target advertisement according to the query information input by the query user, the relevance between the recommendation information and the query information input by the query user can be ensured to be stronger. Meanwhile, in the process of information analysis and generation of the recommendation information generation model corresponding to the cluster type, user interest information corresponding to the query user is introduced, so that the generated recommendation information has better individuation, better accords with the query preference of the user, and further improves the click rate of the user on the target advertisement.
In some embodiments of the present disclosure, in the pre-completed data processing, the data involved includes, but is not limited to, answer data of advertisements and user behavior data in historical data. The answer data of the advertisement can comprise a description document of the advertisement product, the description document comprises common questions of the advertisement product and answers corresponding to the common questions, and the answer data of the advertisement can also comprise a question and answer document which is sorted out by an operator according to user feedback, market research and other conditions. In some other embodiments, the response data for the advertisement may also include user-written questions for the merchant and user-written comment response content for the merchant.
The user behavior data in the historical data refers to user historical search operation recorded in a database, and mainly comprises query information input by the user history, click conditions of recommended advertisements in search results after the query information is input, and recommendation information and advertisement jargon information corresponding to the clicked advertisements.
Fig. 2 is a schematic flow chart of a data formatting method according to an embodiment of the present disclosure, in which before clustering the advertisement identifiers according to the question-answer data of the advertisements to obtain cluster types, the method further includes steps S10 to S20.
Step S10: and formatting the question and answer data of the advertisement to obtain the question and answer data with a preset question and answer data format.
And uniformly arranging the formats of the question and answer data of each advertisement, and arranging all the question and answer data into a uniform preset question and answer data format. For example, a description document of a general question of an advertised product and a question and answer document collated by an operator may be collated in the form of < advertisement identification mark, question data, description/answer data >.
Step S20: and carrying out format arrangement on the user behavior data of the user to obtain the user behavior data with a preset behavior data format.
And uniformly sorting the user behavior data of each user in the historical data in format, and sorting all the user behavior data into a uniform preset behavior data format. For example, the user behavior data may be arranged in the form of < user identification flag, advertisement identification flag, historical query information, historical advertisement recommendation information >.
In this embodiment, by optimizing the data set formats of the question and answer data and the user behavior data, data processing of the question and answer data of the advertisement and the user behavior data of the user is facilitated, so that optimization of the recommendation information generation process of the advertisement is further facilitated.
In one embodiment, the description document of the common questions of the advertisement products and the question and answer document sorted by the operator can be sorted into the question and answer data document of the advertisement. And the data in the question-answer data document are question-answer data with a preset question-answer data format. When the advertisement identification marks are clustered in advance according to the question and answer data of the advertisements, the advertisement answer data (namely the third column of data in the question and answer data document with the preset question and answer data format) is used as a clustering basis, the advertisement identification marks corresponding to the answer data meeting the preset similar conditions are aggregated into a cluster, and different cluster types are obtained. The process of determining the answer data meeting the preset similar condition may be that when the distance between the two answer data vectors is smaller than a preset value, the two answer data vectors are determined to meet the preset similar condition, and the advertisement identification marks corresponding to the two answer data vectors are aggregated into a cluster.
Cluster analysis refers to an analysis process that groups advertisement identification into multiple classes composed of similar objects, where objects in the same cluster have great similarity and objects between different clusters have great dissimilarity. In the present embodiment, advertisement identifiers are grouped into a plurality of clusters consisting of advertisement identifiers having similar answer data by comparing similarities between answer data corresponding to the advertisement identifiers. In some embodiments of the present disclosure, advertising identifying indicia having similar answer data may be aggregated into a cluster using clustering algorithms including, but not limited to, distance-based clustering algorithms, density-based clustering algorithms, and word vector-based clustering algorithms to obtain different said cluster types.
After the cluster analysis of the advertisements is completed, data pairs corresponding to all advertisement identification marks in the formed cluster types (namely, the second column data and the third column data in the question-answer data document with the preset question-answer data format) can be used as a fine-grained data set, and when a recommendation information generation model is independently constructed for each cluster type, the recommendation information generation model constructed correspondingly is trained by using the fine-grained data set of each cluster type.
Fig. 3 is a flowchart illustrating a method for aggregating advertisement identifiers with similar answer data according to an embodiment of the present disclosure, in which in one embodiment, the advertisement identifiers corresponding to answer data meeting a preset similar condition are aggregated into a cluster, and acquiring different cluster types includes the following steps S30 to S40.
Step S30: and converting the answer data into an answer data vector with a fixed dimension by adopting a paragraph vector method.
And classifying the third column of data (namely answer data) in the question and answer data with the preset question and answer data format according to the first column of data (namely the advertisement identification marks), and classifying and sorting the answer data corresponding to the advertisements with the same advertisement identification marks to obtain an answer data document.
And converting the answer data into an answer data vector with a fixed dimension. In this embodiment, the answer data document of the advertisement may be converted into a fixed-dimension vector using the doc2vec model existing in the industry. The doc2vec model can convert text contents in answer data documents into a vector with a fixed dimension in a K-dimensional vector space through training. Further, the similarity of the answer data vector in the vector space may be used to represent the semantic similarity of the text of the answer data.
Step S40: and clustering the answer data vectors by adopting a K-means clustering algorithm, and clustering the advertisements into clusters with a preset number.
In some embodiments of the present disclosure, the answer data vectors may be clustered using the K-Means method. The K-Means Clustering Algorithm (K-Means Clustering Algorithm) is a Clustering analysis Algorithm for iterative solution. Dividing the answer data vector into k groups (wherein k is a natural number larger than zero), and randomly selecting k objects as initial clustering centers. The distance between each object and the respective cluster center is calculated, and each object is assigned to the cluster center closest to it, the cluster centers and the objects assigned to them representing a cluster. The cluster generated by clustering is a collection of a set of data objects that are similar to objects in the same cluster and distinct from objects in other clusters. The cluster center of a cluster is recalculated for each sample assigned based on the objects existing in the cluster. This process is repeated until some termination condition is met, thereby dividing the collection of physical or abstract objects into classes composed of similar objects.
After clustering processing is carried out on the advertisement identification marks according to the question and answer data of the advertisements by using a K-Means method, a plurality of cluster types are obtained, and each cluster type comprises a plurality of advertisement identification marks of the advertisements, so that a mapping relation < advertisement identification mark, cluster type > can be constructed.
By optimizing the question-answer data format of the advertisement, clustering the advertisement identification marks according to the question-answer data of the advertisement, aggregating the advertisement identification marks with similar answer data into a cluster, acquiring different cluster types, and independently constructing a recommendation information generation model for each cluster type, the problem that recommendation information output by a single recommendation information generation model is irrelevant to input query information can be solved, the loss of information characteristics is reduced, and the effect that the generated recommendation information is more appropriate is achieved.
In one embodiment, before clustering the answer data vectors by using a K-means clustering algorithm and clustering the advertisements into a preset number of clusters, the method further includes setting an initial value of the preset number. When the answer data vectors are clustered by adopting a K-Means clustering algorithm, the initial value of the clustering number K can be preset according to personal experience, and advertisements are clustered into K cluster types by adopting a K-Means method. In some other embodiments, the initial value of the cluster number k may also be set according to the type number of the advertisement predicted by the operator.
In one embodiment, after the advertisements are clustered into the clusters with the preset number by using the K-means clustering algorithm, the method further includes adjusting the values of the preset number according to the clustering effect. And judging the correlation among the advertisements in the k cluster types obtained by clustering according to experience by an operator to judge whether the value of the clustering number k is proper or not. If the correlation among the advertisements in the k cluster types is poor, the value of k is adjusted, then clustering operation is carried out again, the process of adjusting clustering is repeatedly executed to select the k value with the optimal clustering effect, the correlation among the advertisements in the k cluster types is ensured, and the effectiveness of the recommendation information generation model corresponding to the cluster types is improved.
In one embodiment, the user behavior data comprises user identification marks and historical query information, the generation of the user interest information according to the user behavior data of the user in advance comprises the steps of using the user behavior data as training data of a word vector model, and generating the user interest information, wherein one user corresponds to one user interest information. By recording data for each query operation of the user, user behavior data of the user can be acquired.
For example, a user identified as a inputs "what nutrition needs to be supplemented by a baby for three months? "and after reading the recommendation information of the related advertisement product, clicking the advertisement with the product identification mark B for viewing. Recording the above operation of user a, one can obtain a < a, B, "what nutrition should be supplemented by baby for three months? ", B corresponds to recommendation information >. By researching the query information input by the user, the advertisement clicked by the user, the recommendation information corresponding to the advertisement and other data, the preference of the user can be determined, and the recommendation information of the advertisement is optimized according to the interest preference of the user.
In some embodiments of the present disclosure, the doc2vec model may also be used to generate the user interest information from the first column of data and the third column of data (i.e., the user identification and the historical query information) in the preset behavior data format in the historical data. The doc2vec model can also convert text contents in user behavior data into vectors in a K-dimensional vector space through training. The vector of each user can be generated by arranging the user behavior data into the following format < user identification, word segmentation result of all historical query information > as the training data of the doc2vec model. The vector of each user is used as the user interest information of each user, and one user corresponds to one user interest information to obtain an interest vector dictionary < I1, I2, …, Iu >.
By constructing the user interest dictionary, when the recommendation information is generated by the recommendation information generation model, the generated recommendation information is more personalized and more in line with the preference of the user by combining the user interest information corresponding to the user, so that the recommendation effect of the recommendation information on the target advertisement is optimized.
Fig. 4 is a flowchart illustrating a method for constructing different recommendation information generation models according to an embodiment of the present disclosure, where the user behavior data includes a user identifier, an advertisement identifier, historical query information, and historical advertisement recommendation information, and the different recommendation information generation models that are constructed in advance for different cluster types include the following steps S50 to S60.
Step S50: the method comprises the steps of obtaining an advertisement identification mark corresponding to a target cluster type, obtaining user behavior data corresponding to the target cluster type according to the advertisement identification mark, and obtaining user interest information corresponding to the user identification mark according to the user identification mark in the user behavior data.
In the pre-completed data processing, different recommendation information generation models may be constructed for different cluster types. And when a recommendation information generation model corresponding to the cluster type is established aiming at the target cluster type, determining the advertisement identification mark contained in the target cluster type according to the mapping relation between the cluster type and the advertisement identification mark. Because the user behavior data comprises the advertisement identification mark, the user behavior data which can be used for training the recommendation information generation model constructed by the target cluster type can be determined according to the advertisement identification mark corresponding to the target cluster type. And user interest information corresponding to the user can be acquired according to the user identification in the user behavior data.
For example, the first cluster type mainly includes advertisement identification marks of various maternal and infant products, the user behavior data of the first user and the second user in the historical data includes the advertisement identification marks of the maternal and infant products, and the user behavior data corresponding to the advertisement identification marks of the maternal and infant products included by the first user and the second user can be used as training data of a recommendation information generation model corresponding to the first cluster type. According to the user identification marks of the first user and the second user, user interest information of the first user and the second user can be obtained.
Step S60: and training the information generation model according to the user interest information corresponding to the user identification and the historical query information in the user behavior data corresponding to the target cluster type to obtain the recommendation information generation model corresponding to the target cluster type.
In the present embodiment, a recommendation information generation model corresponding to a cluster type is constructed using a Seq2Seq model as an information generation model. And taking the user interest information corresponding to the user identification and the historical query information in the user behavior data corresponding to the cluster type as training data, and training the Seq2Seq model to obtain a recommendation information generation model corresponding to the cluster type. And training recommendation information generation models which are correspondingly constructed for the k cluster types respectively, wherein the training data are trained by adopting a proper training frame, wherein each cluster type comprises user behavior data < user identification, advertisement identification, historical query information and historical advertisement recommendation information > corresponding to the advertisement identification. In one embodiment, a deep learning framework such as TensorFlow, Keras, PyTorch, Caffe, deep learning4j, etc. may be used.
For example, the first cluster type mainly comprises advertisement identification marks of various maternal and infant products, user behavior data containing the advertisement identification marks of the maternal and infant products can be used as training data of the first cluster type, a recommendation information generation model corresponding to the first cluster type is optimized, and it is ensured that recommendation information which is more relevant and targeted to the maternal and infant product advertisements can be constructed when the recommendation information generation model corresponding to the first cluster type generates recommendation information according to query information input by a user. Meanwhile, when the recommendation information generation model corresponding to the first cluster type is trained, user interest information of the first user and user interest information of the second user can be introduced, so that the recommendation effect of the recommendation information generation model corresponding to the first cluster type is personalized and better accords with the interest preference of the users.
In one embodiment, the standard Seq2Seq model comprises three parts: encoder, Encoder Vector, and Decoder. The Encoder is formed by stacking a plurality of cyclic units (Long Short-Term Memory network LSTM, Long Short-Term Memory or Gate cyclic Unit GRU, Gate Recurrent Unit), and each cyclic Unit receives a single element in an input sequence and propagates forwards. The Encode Vector is the output of the last layer of the Encode, and has the function of capturing the meaning of the whole input and is used as the initial input of the Decode. The Decoder is also formed by stacking a plurality of cyclic units, each cyclic unit takes the hidden state of the previous cyclic unit as input, and each cyclic unit predicts the output y in each step until the stop symbol is output.
According to the information recommending method, when different recommended information generation models corresponding to different cluster types are constructed, the input part of the Decoder in the Seq2Seq model is improved, in order to enable the input characteristic of the Decoder to be more effective, user interest information corresponding to a user is introduced into the input part of the Decoder to serve as personalized Decoder input characteristics, the recommended information generated by the improved Seq2Seq model is enabled to be better personalized and more in line with the interest of the user, and therefore the recommending effect is improved.
Fig. 5 is a flowchart illustrating a method for obtaining a recommendation information generation model corresponding to a target cluster type according to an embodiment of the present disclosure, where in one embodiment, training the information generation model according to user interest information corresponding to a user identification and historical query information in user behavior data corresponding to a cluster type to obtain the recommendation information generation model corresponding to the target cluster type includes the following steps S61 to S69.
Step S61: and taking historical query information in the user behavior data corresponding to the target cluster type as the input of an encoder in the information generation model to acquire encoded output information.
Each cluster type comprises one or more advertisement identification marks, the user behavior data also comprises the advertisement identification marks, the user behavior data comprising the advertisement identification marks corresponding to the cluster types in the data is used as training data, and the information generation model is trained to obtain the recommendation information generation model corresponding to the cluster type. In the present embodiment, a Seq2Seq model is adopted as the information generation model. When the Seq2Seq model is trained by using the training data, historical query information in the user behavior data is used as input of an encoder in the information generation model.
For example, the first cluster type mainly includes advertisement identification marks of various maternal and infant products, and meanwhile, the user behavior data of the first user in the historical data includes the advertisement identification marks of the maternal and infant products. When the first cluster type is used as a target cluster type to construct a corresponding recommendation information generation model, user behavior data related to a first user can be used as training data of a Seq2Seq model, and the recommendation information generation model corresponding to the first cluster type is generated in an optimized mode. First historical query information in user behavior data of a first user is used as input of an encoder in a Seq2Seq model. The encoder of the Seq2Seq model will perform data processing on the first historical query information to generate first encoded output information.
Step S63: and calculating the similarity information of the user interest information corresponding to the user identification and the coding output information.
And an encoder in the information generation model performs data processing on historical query information in the user behavior data to acquire encoded output information EncderVec. Meanwhile, corresponding user interest information I can be obtained according to the user identification marks in the user behavior datak. Calculating user interest information IkSimilarity information s with encoded output information EncderVeck. Wherein the similarity information skThe calculation method of (2) is as follows:
Figure BDA0003192436200000151
where sim (x, y) represents the similarity of the computed vector x and the vector y, where x isiRefers to the value of the ith dimension, y, of the x vectoriRefers to the value of the ith dimension of the y vector. Calculating the similarity information s between the user interest information and the coding output informationkIn time, x can be made to be user interest information IkAnd y is the encoded output information EncderVec.
For example, first user interest information of the first user may be determined from a user identification of the first user. The first similarity information may be obtained by performing similarity calculation on the first user interest information and the first encoded output information.
Step S65: and splicing the similarity information, the coding output information and the user interest information to obtain spliced information.
For similarity information skEncderVec and IkAnd splicing to obtain splicing information. Similarity data skSplicing the user interest vector I before EncodeVeckAfter splicing to EncodeVec, the splicing information is sk-Encoder Vector-Ik。
For example, the first splicing information may be obtained by splicing the first similarity information, the first encoding output information, and the first user interest information. The data format of the first splicing information is the first similarity information, the first coding output information and the first user interest information.
Step S67: and taking the splicing information as the input of a decoder in the information generation model to acquire training recommendation information.
Will splice information sk-Encoder Vector-IkAs input to a decoder in an information generation model, the decoderAnd after the input information is processed, corresponding text information is output. The text information output by the decoder is training recommendation information generated by the information generation model.
For example, the first splicing information is used as an input of a Decoder in the Seq2Seq model, and an output of the Decoder is the first training recommendation information.
Step S69: and comparing the training recommendation information with the historical advertisement recommendation information, and optimizing the information generation model according to the comparison result to obtain a recommendation information generation model corresponding to the target cluster type.
The user behavior data also comprises historical advertisement recommendation information corresponding to the historical query information, training recommendation information generated by the information generation model is compared with the historical advertisement recommendation information corresponding to the historical query information, and whether the text generation effect of the information generation model is expected or not is judged. And if the training recommendation information does not meet the expectation, performing corresponding improvement optimization on the information generation model until the training recommendation information generated by the improved information generation model has high similarity with the historical advertisement recommendation information. And defining the improved information generation model with better recommendation information generation effect as a recommendation information generation model corresponding to the target cluster type.
For example, the user behavior data of the first user includes historical advertisement recommendation information of the historical query information corresponding to the first user click. And comparing the first training recommendation information with historical advertisement recommendation information corresponding to the advertisement clicked by the first user, and judging whether the Seq2Seq model can generate historical advertisement recommendation information with high correlation with the historical query information of the first user. And repeatedly executing the training optimization process of the Seq2Seq model to select a model with the optimal recommendation information generation effect so as to obtain a recommendation information generation model corresponding to the first cluster type. And a recommendation information generation model corresponding to the first cluster type is ensured, and recommendation information of the target advertisement which is more suitable for the query requirement of the user can be generated according to the query information input by the user.
And training recommendation information generation models which are correspondingly constructed for the k cluster types respectively to obtain different recommendation information generation models corresponding to different cluster types. And training different recommendation information generation models corresponding to different cluster types by adopting a proper training frame.
In one embodiment, the cluster type to which the target advertisement belongs is determined according to the advertisement identification of the target advertisement, and the user interest information of the inquiring user is determined according to the user identification of the inquiring user. Suppose that the user interest information corresponding to the user k1 is Ik1The query information input by the user k1 is recorded as query 1. An Encoder Encoder in a recommendation information generation model corresponding to the cluster type converts query information query1 input by a query user into a vector format to obtain a q1 vector, and the q1 vector is used as the last layer of the Encoder to output EncodeVec 1. The implementation method for converting the query information query1 input by the query user into the vector format to obtain the q1 vector includes but is not limited to the steps of adopting a common text vector model, performing word segmentation on all query information, and then adopting word2vec training to obtain a dictionary from words to query information word vectors and recording the dictionary as W2V.
For example, a method for calculating the q1 vector includes performing word segmentation on query information query1 to obtain word list words. And searching a word vector corresponding to each participle in the word list words from the vector dictionary W2V. And solving the average value of the word vectors corresponding to each participle in the word list words so as to obtain the q vector of the query information query 1.
User interest information Ik1The similarity with EncoderVec1 is recorded as similarity data sk1User interest information Ik1Similarity calculation is carried out with EncodeVec 1 to obtain similarity data sk1. Similarity data sk1Splicing to EncodeVec 1, user interest vector Ik1After being spliced into EncodeVec 1, namely the spliced data is sk1-EncoderVec1-Ik1The result s after splicingk1-EncoderVec1-Ik1As an input of the Decoder, recommendation information of the target advertisement is thus output using the Decoder.
The recommendation information generation method can ensure that the recommendation information generation modes of the advertisements in all cluster types after clustering are more similar through clustering. The recommendation information generation model is independently constructed for each cluster type, so that the relevance between the recommendation information generated by the recommendation information generation model and the query information input by the user is stronger, and the situation that the recommendation information generated by the recommendation information generation model is irrelevant to the input is further reduced. After the query user inputs query information and determines the advertisements needing to be pushed to the query user, the cluster type of the target advertisements can be determined according to the advertisement identification marks of the target advertisements, and therefore recommendation information of the target advertisements is generated by using a recommendation information generation model corresponding to the cluster type. Meanwhile, the recommendation information generated by the recommendation information generation model is personalized according to the user interest information corresponding to the query user and is more in line with the preference of the user, so that the recommendation effect of the recommendation information on the target advertisement is optimized.
It should be understood that although the various steps in the flowcharts of fig. 1-5 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 1-5 may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed in turn or alternately with other steps or at least some of the other steps or stages.
Based on the description of the recommendation information generation method described in the above embodiment, the present disclosure also provides a recommendation information generation system. The system may include a system (including a distributed system), software (applications), modules, components, etc. that employ the methods described in embodiments of the specification in conjunction with hardware where necessary to implement such methods. Based on the same innovative concept, the embodiments of the present disclosure provide an apparatus in one or more embodiments as described in the following embodiments. Since the implementation scheme of the apparatus for solving the problem is similar to that of the method, the specific implementation of the apparatus in the embodiment of the present specification may refer to the implementation of the foregoing method, and repeated details are not repeated. As used hereinafter, the term "unit" or "module" may be a combination of software and/or hardware that implements a predetermined function. Although the means described in the embodiments below are preferably implemented in software, an implementation in hardware, or a combination of software and hardware is also possible and contemplated.
Based on the description of the foregoing method embodiment, the present disclosure also provides a recommendation information generation system. Fig. 6 is a schematic structural diagram of a recommendation information generation system according to an embodiment of the present disclosure, and in an embodiment, the recommendation information generation system may include an advertisement cluster identification module 100, a user interest identification module 200, and a recommendation information generation module 300.
The advertisement cluster recognition module 100 is configured to obtain an advertisement recognition identifier of a target advertisement, and determine a cluster type to which the target advertisement belongs according to the advertisement recognition identifier, where the cluster type is obtained by clustering the advertisement recognition identifier in advance according to question and answer data of the advertisement, and different cluster types correspond to different pre-constructed recommendation information generation models. A user interest identification module 200, configured to obtain a user identification of a querying user, and determine user interest information of the querying user according to the user identification; the user interest information is generated in advance according to user behavior data of the user. The recommendation information generating module 300 is configured to input the user interest information and the query information of the query user into a recommendation information generating model corresponding to the cluster type to obtain recommendation information of the target advertisement.
With regard to the system in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.
It is to be understood that the various embodiments of the methods, apparatus, etc. described above are described in a progressive manner, and like/similar elements may be referred to one another, with each embodiment focusing on differences from the other embodiments. Reference may be made to the description of other method embodiments for relevant points.
Fig. 7 is a block diagram illustrating a recommendation information generating device or system S00 according to an example embodiment. Referring to FIG. 7, a recommendation information generating device or system S00 includes a processing component S20 that further includes one or more processors and memory resources, represented by memory S22, for storing instructions, such as applications, executable by the processing component S20. The application program stored in the memory S22 may include one or more modules each corresponding to a set of instructions. Further, the processing component S20 is configured to execute instructions to perform the above-described method.
The recommendation information generating apparatus or system S00 may further include: the power supply component S24 is configured to perform power management of the recommendation information generating device or system S00, the wired or wireless network interface S26 is configured to connect the recommendation information generating device or system S00 to a network, and the input/output (I/O) interface S28. The recommendation information generating device or system S00 may be operable based on an operating system stored in the memory S22, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, or the like.
In an exemplary embodiment, a computer-readable storage medium comprising instructions, such as the memory S22 comprising instructions, executable by the processor of the recommendation information generating device or system S00 to perform the above method is also provided. The storage medium may be a computer-readable storage medium, which may be, for example, a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.
In an exemplary embodiment, a computer program product is also provided that includes instructions executable by a processor of the recommendation information generating device or system S00 to perform the above-described method.
In the description herein, references to the description of "some embodiments," "other embodiments," "desired embodiments," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, a schematic description of the above terminology may not necessarily refer to the same embodiment or example.
The technical features of the embodiments described above may be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the embodiments described above are not described, but should be considered as being within the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims (11)

1. A recommendation information generation method, comprising:
acquiring an advertisement identification mark of a target advertisement, and determining a cluster type to which the target advertisement belongs according to the advertisement identification mark, wherein the cluster type is obtained by clustering the advertisement identification mark according to question and answer data of the advertisement in advance, and different cluster types correspond to different pre-constructed recommendation information generation models;
acquiring a user identification of a query user, and determining user interest information of the query user according to the user identification; the user interest information is generated in advance according to user behavior data of a user;
and inputting the user interest information and the query information of the query user into a recommendation information generation model corresponding to the cluster type to acquire the recommendation information of the target advertisement.
2. The recommendation information generating method according to claim 1, wherein before clustering the advertisement identification tag according to question-answer data of an advertisement to obtain a cluster type, the method further comprises:
formatting the question and answer data of the advertisement to obtain the question and answer data with a preset question and answer data format;
and carrying out format arrangement on the user behavior data of the user to obtain the user behavior data with a preset behavior data format.
3. The recommendation information generation method according to claim 1, wherein the question-answer data includes question data and answer data, and the clustering the advertisement identification marks in advance according to the question-answer data of the advertisements includes:
and aggregating the advertisement identification marks corresponding to the answer data meeting the preset similar conditions into a cluster to obtain different cluster types.
4. The recommendation information generating method according to claim 3, wherein the aggregating the advertisement identification tags corresponding to the answer data meeting the preset similar condition into a cluster, and acquiring different cluster types comprises:
converting the answer data into an answer data vector with a fixed dimension by adopting a paragraph vector method;
and clustering the answer data vectors by adopting a K-means clustering algorithm, and clustering the advertisements into clusters with a preset number.
5. The recommendation information generating method according to claim 4, wherein before clustering the answer data vectors by using a K-means clustering algorithm and clustering the advertisements into a preset number of clusters, the method further comprises:
setting an initial value of the preset number;
after the advertisements are clustered into a preset number of clusters by adopting a K-means clustering algorithm, the method further comprises the following steps:
and adjusting the value of the preset number according to the clustering effect.
6. The recommendation information generation method according to claim 1, wherein the user behavior data includes a user identification and historical query information, and generating user interest information in advance according to the user behavior data of the user includes:
and taking the user behavior data as training data of a word vector model to generate user interest information, wherein one user corresponds to one user interest information.
7. The recommendation information generation method according to claim 1, wherein the user behavior data includes a user identification identifier, an advertisement identification identifier, historical query information, and historical advertisement recommendation information, and the different cluster types correspond to different pre-constructed recommendation information generation models including:
acquiring an advertisement identification mark corresponding to the target cluster type, acquiring user behavior data corresponding to the target cluster type according to the advertisement identification mark, and acquiring user interest information corresponding to the user identification mark according to the user identification mark in the user behavior data;
and training the information generation model according to the user interest information corresponding to the user identification and the historical query information in the user behavior data corresponding to the target cluster type to obtain the recommendation information generation model corresponding to the target cluster type.
8. The method for generating recommendation information according to claim 7, wherein the training an information generation model according to the user interest information corresponding to the user identification and the historical query information in the user behavior data corresponding to the target cluster type to obtain the recommendation information generation model corresponding to the target cluster type comprises:
taking historical query information in user behavior data corresponding to the target cluster type as input of an encoder in an information generation model, and acquiring encoding output information;
calculating the similarity information between the user interest information corresponding to the user identification and the coded output information;
splicing the similarity information, the coding output information and the user interest information to obtain spliced information;
taking the splicing information as the input of a decoder in the information generation model to acquire training recommendation information;
and comparing the training recommendation information with historical advertisement recommendation information, and optimizing the information generation model according to the comparison result to obtain a recommendation information generation model corresponding to the target cluster type.
9. A recommendation information generation system, comprising:
the advertisement cluster identification module is used for acquiring an advertisement identification mark of a target advertisement, and determining a cluster type to which the target advertisement belongs according to the advertisement identification mark, wherein the cluster type is obtained by clustering the advertisement identification mark according to question and answer data of the advertisement in advance, and different cluster types correspond to different pre-constructed recommendation information generation models;
the user interest identification module is used for acquiring a user identification mark of a query user and determining user interest information of the query user according to the user identification mark; the user interest information is generated in advance according to user behavior data of a user;
and the recommendation information generation module is used for inputting the user interest information and the query information of the query user into a recommendation information generation model corresponding to the cluster type so as to obtain the recommendation information of the target advertisement.
10. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor realizes the steps of the method of any one of claims 1 to 8 when executing the computer program.
11. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 8.
CN202110882147.8A 2021-08-02 2021-08-02 Recommendation information generation method, system, computer device and storage medium Pending CN113672804A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110882147.8A CN113672804A (en) 2021-08-02 2021-08-02 Recommendation information generation method, system, computer device and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110882147.8A CN113672804A (en) 2021-08-02 2021-08-02 Recommendation information generation method, system, computer device and storage medium

Publications (1)

Publication Number Publication Date
CN113672804A true CN113672804A (en) 2021-11-19

Family

ID=78541171

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110882147.8A Pending CN113672804A (en) 2021-08-02 2021-08-02 Recommendation information generation method, system, computer device and storage medium

Country Status (1)

Country Link
CN (1) CN113672804A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116823410A (en) * 2023-08-29 2023-09-29 阿里巴巴(成都)软件技术有限公司 Data processing method, object processing method, recommending method and computing device

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104063383A (en) * 2013-03-19 2014-09-24 北京三星通信技术研究有限公司 Information recommendation method and device
CN108520444A (en) * 2018-04-12 2018-09-11 中国平安人寿保险股份有限公司 Insurance products recommend method, unit and computer readable storage medium
CN110263242A (en) * 2019-01-04 2019-09-20 腾讯科技（深圳）有限公司 Content recommendation method, device, computer readable storage medium and computer equipment
CN110457452A (en) * 2019-07-08 2019-11-15 汉海信息技术(上海)有限公司 Rationale for the recommendation generation method, device, electronic equipment and readable storage medium storing program for executing
CN111882370A (en) * 2020-09-27 2020-11-03 武汉卓尔数字传媒科技有限公司 Advertisement recommendation method and device and electronic equipment
WO2021023249A1 (en) * 2019-08-06 2021-02-11 北京三快在线科技有限公司 Generation of recommendation reason

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104063383A (en) * 2013-03-19 2014-09-24 北京三星通信技术研究有限公司 Information recommendation method and device
CN108520444A (en) * 2018-04-12 2018-09-11 中国平安人寿保险股份有限公司 Insurance products recommend method, unit and computer readable storage medium
CN110263242A (en) * 2019-01-04 2019-09-20 腾讯科技（深圳）有限公司 Content recommendation method, device, computer readable storage medium and computer equipment
CN110457452A (en) * 2019-07-08 2019-11-15 汉海信息技术(上海)有限公司 Rationale for the recommendation generation method, device, electronic equipment and readable storage medium storing program for executing
WO2021023249A1 (en) * 2019-08-06 2021-02-11 北京三快在线科技有限公司 Generation of recommendation reason
CN111882370A (en) * 2020-09-27 2020-11-03 武汉卓尔数字传媒科技有限公司 Advertisement recommendation method and device and electronic equipment

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
BARCELONA: "Topic-driven Ensemble for Online Advertising Generation", PROCEEDINGS OF THE 28TH INTERNATIONAL CONFERENCE ON COMPUTATIONAL LINGUISTICS, LINGUISTICS，BARCELONA, SPAIN (ONLINE), vol. 28, no. 2273, pages 1 - 11 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116823410A (en) * 2023-08-29 2023-09-29 阿里巴巴(成都)软件技术有限公司 Data processing method, object processing method, recommending method and computing device
CN116823410B (en) * 2023-08-29 2024-01-12 阿里巴巴(成都)软件技术有限公司 Data processing method, object processing method, recommending method and computing device

Similar Documents

Publication Publication Date Title
CN111753060B (en) Information retrieval method, apparatus, device and computer readable storage medium
CN110162593B (en) Search result processing and similarity model training method and device
CN111859960B (en) Semantic matching method, device, computer equipment and medium based on knowledge distillation
CN112119388A (en) Training image embedding model and text embedding model
CN112800170A (en) Question matching method and device and question reply method and device
CN112434151A (en) Patent recommendation method and device, computer equipment and storage medium
CN112395487B (en) Information recommendation method and device, computer readable storage medium and electronic equipment
CN111046275A (en) User label determining method and device based on artificial intelligence and storage medium
US20220172260A1 (en) Method, apparatus, storage medium, and device for generating user profile
CN112074828A (en) Training image embedding model and text embedding model
CN111831924A (en) Content recommendation method, device, equipment and readable storage medium
CN112836509A (en) Expert system knowledge base construction method and system
Jagabathula et al. A model-based embedding technique for segmenting customers
CN110866102A (en) Search processing method
CN113011172A (en) Text processing method and device, computer equipment and storage medium
CN113821527A (en) Hash code generation method and device, computer equipment and storage medium
CN113282729A (en) Question-answering method and device based on knowledge graph
CN115809887A (en) Method and device for determining main business range of enterprise based on invoice data
CN112597292B (en) Question reply recommendation method, device, computer equipment and storage medium
CN113672804A (en) Recommendation information generation method, system, computer device and storage medium
CN114329004A (en) Digital fingerprint generation method, digital fingerprint generation device, data push method, data push device and storage medium
CN113327132A (en) Multimedia recommendation method, device, equipment and storage medium
CN112565903B (en) Video recommendation method and device, server and storage medium
CN111104422A (en) Training method, device, equipment and storage medium of data recommendation model
CN115455152A (en) Writing material recommendation method and device, electronic equipment and storage medium

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