CN113239182A - Article recommendation method and device, computer equipment and storage medium - Google Patents
Article recommendation method and device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN113239182A CN113239182A CN202110545897.6A CN202110545897A CN113239182A CN 113239182 A CN113239182 A CN 113239182A CN 202110545897 A CN202110545897 A CN 202110545897A CN 113239182 A CN113239182 A CN 113239182A
- Authority
- CN
- China
- Prior art keywords
- article
- index value
- reading
- user
- recommended
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/335—Filtering based on additional data, e.g. user or group profiles
-
- 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
-
- 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
Abstract
The invention relates to the technical field of information service, and discloses an article recommendation method, a device, computer equipment and a storage medium, namely, an article hotspot index value of an article to be recommended is firstly obtained, an article characteristic index value, a user recommendation refusing index value and a user reading index value of the article to be recommended on a target user are obtained, then a pentagonal average weighting algorithm is adopted to calculate and obtain a comprehensive recommendation index value of the article to be recommended from five angles of article hotspots, article characteristics, user recommendation refusing behaviors, user reading behaviors and the like, and finally whether the article to be recommended is pushed is decided according to a comparison result of the comprehensive recommendation index value and a threshold value, so that whether the target user is interested in the article to be recommended can be predicted, and the purpose of comprehensively expressing user behaviors and information attributes of the article is realized, and the recommendation method is further helpful for the recommender to suggest more interesting articles to the target user, and the article pushing accuracy is improved.
Description
Technical Field
The invention belongs to the technical field of information service, and particularly relates to an article recommendation method and device, computer equipment and a storage medium.
Background
Nowadays, the amount of information available on the internet and the number of users are explosively increasing, resulting in problems of information overload and not enabling users to timely and effectively access interesting content on the internet, which makes the recommendation system a popular and important research topic. The recommendation system filters out decisive information segments from a large amount of dynamically generated information by collecting data such as interests, preferences and past decisions on different things of a user, so that the problem of information overload is solved. By utilizing different filtering and recommendation methods, the recommendation system can predict whether a target user is interested in something based on the user's profile. Not only can the recommendation system make it more efficient for users to find interesting things, but also service providers can actively send items to users to improve their quality of service and user experience.
A recommendation system is defined as a decision-making method for recommending personalized, proprietary content and services to a user that are relevant to the user's interests and preferences in order to solve the information overload problem described above. The purpose of a recommendation system is to generate meaningful and interesting suggestions on content, items and/or products, etc. for a group of users. Generally, the recommendation problem is reduced to a problem of interest level estimation for items that the user has not seen yet. Intuitively, the estimation is typically based on the ranking given to other items by the user or similar users, thereby ensuring that the recommendation system can recommend to the user what they have not seen and are meaningful, such as items, and content, based on the estimated item rating. Examples of such applications include, but are not limited to, recommending shoes, clothing, and other products on an e-commerce web site, short videos and movies on an online video web site, and news on the top of today, among others.
Currently, researchers have developed various methods of creating recommendation systems, i.e., based on the key technology to build them, recommendation systems can be roughly divided into three categories: (1) collaborative filtering, namely recommending articles liked by people with similar likes and likes to a user in a collaborative filtering system; (2) content-based recommendations, i.e. items recommended to the user are similar to items liked by the user in the past; (3) hybrid approaches, i.e. approaches that combine the aforementioned collaborative approaches and content-based recommendations. However, there are many areas in the prior art that can be improved, one of which is: how to provide a better way to represent user behavior and information about items to be recommended in order to achieve a more accurate recommendation goal.
Disclosure of Invention
In order to better represent user behaviors and information related to items to be recommended and solve the problem that articles are difficult to accurately push currently, the invention aims to provide a novel article recommendation method, a novel article recommendation device, a novel computer device and a novel storage medium, which can comprehensively express user behaviors and information attributes of the articles from five aspects of article hotspots, article characteristics, user recommendation refusing behaviors and user reading behaviors and the like to obtain a comprehensive recommendation index value of the articles to be recommended, thereby being beneficial to a recommender to suggest more interesting articles to a target user, improving the article pushing accuracy and facilitating practical application and popularization.
In a first aspect, the present invention provides an article recommendation method, including:
obtaining an article hotspot index value of an article to be recommended, obtaining an article characteristic index value, a user recommendation refusing index value and a user reading index value of the article to be recommended on a target user, wherein the article hotspot index value is used for reflecting the similarity degree of the article to be recommended and a real-time hotspot news article, the article characteristic index value is used for reflecting the similarity degree of the article to be recommended and the prior reading article of the target user, the user characteristic index value is used for reflecting how much the similar users of the target user read the article to be recommended, the user rejections index value is used for reflecting the similarity degree of the article to be recommended and the prior rejections article of the target user, the user reading index value is used for reflecting the behavior depth of the target user when reading the previous similar articles of the articles to be recommended;
according to the article hotspot index value, the article characteristic index value, the user recommendation refusing index value and the user reading index value, calculating a comprehensive recommendation index value A of the article to be recommended according to the following formulaP：
AP＝w1*A1+w2*A2+w3*A3+w4*A4+w5*A5
In the formula, w1Representing a preset weight coefficient, A, corresponding to the article hotspot index value1Representing the article hotspot index value, w2Representing a preset weight coefficient, A, corresponding to the article feature index value2Representing the value of the article characteristic index, w3Representing a preset weight coefficient corresponding to said user characteristic index value, A3Representing said user characteristic index value, w4Representing a preset weight coefficient corresponding to the user recommendation rejection index value, A4Represents the user recommendation refusal index value w5Representing a preset weighting factor, A, corresponding to said user reading index value5Representing the user reading index value;
when the comprehensive recommended index value A is judgedPAnd when the recommendation index is larger than or equal to a preset recommendation index threshold value, pushing the article to be recommended to the target user.
Based on the content of the invention, an article recommendation scheme adopting a pentagonal average weighting algorithm can be provided, namely, an article hotspot index value of an article to be recommended is obtained, an article characteristic index value, a user recommendation refusal index value and a user reading index value of the article to be recommended on a target user are obtained, then a pentagonal average weighting algorithm is adopted to calculate a comprehensive recommendation index value of the article to be recommended from five angles of article hotspots, article characteristics, user recommendation refusal behaviors, user reading behaviors and the like, and finally whether the article to be recommended is pushed is decided according to a comparison result of the comprehensive recommendation index value and a threshold value, so that whether the target user is interested in predicting the article to be recommended can be predicted, the purpose of comprehensively expressing user behaviors and information attributes of the article can be realized, and the problem of insufficient independent images in a traditional recommendation system can be solved, and then help the recommender to suggest more interesting articles to the target user, promote the accuracy of article propelling movement, be convenient for practical application and popularization.
In one possible design, obtaining article hotspot index values of articles to be recommended comprises:
acquiring at least one hot news article in real time;
for each hot news article in the article to be recommended and the at least one hot news article, importing the corresponding article content into a trained neural network model, and obtaining a corresponding article characteristic value;
and calculating the similarity between the article to be recommended and each hot news article according to a cosine similarity method of the following formula:
wherein i represents a positive integer,
representing the similarity between the article to be recommended and the ith hot news article in the at least one hot news article, fPAn article feature value representing the article to be recommended,
an article feature value representing the ith hot news article;
and according to the maximum similarity obtained by calculation and the preset corresponding relation between the multiple pairs of similarity intervals and the numerical value, taking the numerical value corresponding to the similarity interval where the maximum similarity exists as the article hotspot index value.
In one possible design, obtaining an article feature index value of the article to be recommended on a target user includes:
acquiring all prior reading articles of the target user;
for the article to be recommended and each of the all the previously read articles, importing the corresponding article content into a trained neural network model, and acquiring a corresponding article characteristic value;
and calculating the similarity between the article to be recommended and each of the prior reading articles according to a cosine similarity method of the following formula:
wherein j represents a positive integer,
representing the similarity of the article to be recommended and the jth prior reading article in all the prior reading articles, fPAn article feature value representing the article to be recommended,
an article feature value representing the jth prior-read article;
and taking the numerical value corresponding to the similarity interval with the maximum similarity as the article characteristic index value according to the maximum similarity obtained by calculation and the preset corresponding relation between the multiple pairs of similarity intervals and the numerical value.
In one possible design, obtaining a user characteristic index value of the article to be recommended on a target user includes:
acquiring personal data of the target user and personal data of all other users;
aiming at the target user and each other user in all other users, importing the corresponding personal data into the trained neural network model to obtain the corresponding user characteristic value;
and calculating the similarity between the target user and each other user according to a cosine similarity method of the following formula:
in the formula, k represents a positive integer,
representing the similarity of the target user to the kth other user among the all other users, fUA user characteristic value representing the target user,
a user characteristic value representing the kth other user;
screening out a plurality of other users with the former similarity according to the calculation result;
and counting the number of users reading the article to be recommended or the user number ratio among the other users, and taking the number of users or the user number ratio as the user characteristic index value.
In one possible design, obtaining a user recommendation rejection index value of the article to be recommended on a target user includes:
acquiring all prior recommendation refusing articles of the target user, wherein the prior recommendation refusing articles refer to prior recommendation articles which are not read by the target user;
for the article to be recommended and each of the articles rejected earlier, importing the corresponding article content into a trained neural network model to obtain a corresponding article characteristic value;
calculating the similarity between the article to be recommended and each of the prior recommended refusing articles according to a cosine similarity method of the following formula:
wherein, x represents a positive integer,
representing a similarity of the article to be recommended and an xth preceding recommended article among the all preceding recommended articles, fPAn article feature value representing the article to be recommended,
an article characteristic value representing the xth prior rejector article;
and according to the maximum similarity obtained by calculation and the preset corresponding relation between the multiple pairs of similarity intervals and the numerical value, taking the numerical value corresponding to the similarity interval with the maximum similarity as the user recommendation refusing index value.
In one possible design, obtaining a user reading index value of the article to be recommended on a target user includes:
acquiring all prior reading articles of the target user;
for the article to be recommended and each of the all the previously read articles, importing the corresponding article content into a trained neural network model, and acquiring a corresponding article characteristic value;
and calculating the similarity between the article to be recommended and each of the prior reading articles according to a cosine similarity method of the following formula:
wherein y represents a positive integer,
representing the similarity of the article to be recommended and the y-th prior reading article in all the prior reading articles, fPAn article feature value representing the article to be recommended,
an article feature value representing the y-th prior-read article;
screening a plurality of prior reading articles with the former similarity according to the calculation result;
for each of the plurality of prior reading articles, determining a corresponding individual behavior depth value according to corresponding reading behavior acquisition data;
calculating the reading index value A of the user according to the following formula5：
Wherein M represents the number of articles of the plurality of prior reading articles, z represents a positive integer,
a predetermined weight coefficient representing a z-th prior-read article among the plurality of prior-read articles, azAnd the individual behavior depth value of the z th prior reading article in the plurality of prior reading articles is represented, and the preset weight coefficient of the prior reading article is positively correlated with the corresponding similarity or negatively correlated with the corresponding sequence number which is obtained by sequencing from high to low according to the similarity.
In one possible design, for each of the plurality of previous reading articles, determining a corresponding individual behavior depth value according to the corresponding reading behavior collection data includes:
initializing the fast reading behavior index value, the semi-reading behavior index value, the refined reading behavior index value, the like behavior index value, the collection behavior index value, the sharing behavior index value and/or the comment behavior index value to zero;
according to reading behavior collection data corresponding to a previous reading article, updating the fast reading behavior index value, the semi-reading behavior index value, the refined reading behavior index value, the like behavior index value, the collection behavior index value, the sharing behavior index value and/or the comment behavior index value according to the following modes:
when the reading behavior acquisition data shows that the prior reading article is quickly read by the target user, updating the quick reading behavior index value to be a first numerical value larger than zero;
when the reading behavior acquisition data shows that the prior reading article is not read completely by the target user and then quits reading, updating the semi-reading behavior index value to be a second value smaller than zero;
when the reading behavior acquisition data shows that the prior reading article is accurately read by the target user, updating the accurate reading behavior index value to be a third numerical value larger than zero;
when the reading behavior collection data shows that the prior reading article is praised by the target user, updating the praise behavior index value to be a fourth numerical value larger than zero;
when the reading behavior acquisition data show that the prior reading article is collected by the target user, updating the collection behavior index value to be a fifth numerical value larger than zero;
when the reading behavior acquisition data shows that the prior reading article is shared by the target user, updating the sharing behavior index value to be a sixth numerical value larger than zero;
when the reading behavior collection data show that the prior reading article is commented by the target user, updating the comment behavior index value to be a seventh numerical value larger than zero;
taking the average value or the accumulated value of the fast reading behavior index value, the half reading behavior index value, the refined reading behavior index value, the like behavior index value, the collection behavior index value, the sharing behavior index value and/or the comment behavior index value as the individual behavior depth value corresponding to the prior reading article.
In a second aspect, the invention provides an article recommendation device, which comprises an index acquisition module, an index synthesis module and an article pushing module, which are sequentially in communication connection;
the index acquisition module is used for acquiring article hotspot index values of articles to be recommended, acquiring article characteristic index values, user recommendation refusing index values and user reading index values of the articles to be recommended on a target user, wherein the article hotspot index value is used for reflecting the similarity degree of the article to be recommended and a real-time hotspot news article, the article characteristic index value is used for reflecting the similarity degree of the article to be recommended and the prior reading article of the target user, the user characteristic index value is used for reflecting how much the similar users of the target user read the article to be recommended, the user rejections index value is used for reflecting the similarity degree of the article to be recommended and the prior rejections article of the target user, the user reading index value is used for reflecting the behavior depth of the target user when reading the previous similar articles of the articles to be recommended;
the index integration module is used for calculating an integrated recommendation index value A of the article to be recommended according to the article hotspot index value, the article characteristic index value, the user recommendation refusing index value and the user reading index value and according to the following formulaP：
AP＝w1*A1+w2*A2+w3*A3+w4*A4+w5*A5
In the formula, w1Representing a preset weight coefficient, A, corresponding to the article hotspot index value1Representing the article hotspot index value, w2Representing a preset weight coefficient, A, corresponding to the article feature index value2Representing the value of the article characteristic index, w3Representing a preset weight coefficient corresponding to said user characteristic index value, A3Representing said user characteristic index value, w4Representing a preset weight coefficient corresponding to the user recommendation rejection index value, A4Represents the user recommendation refusal index value w5Representing a preset weighting factor, A, corresponding to said user reading index value5Representing the user reading index value;
the article pushing module is used for judging the comprehensive recommended index value APAnd when the recommendation index is larger than or equal to a preset recommendation index threshold value, pushing the article to be recommended to the target user.
In a third aspect, the present invention provides a computer device, comprising a memory and a processor, wherein the memory is used for storing a computer program, and the processor is used for reading the computer program and executing the article recommendation method according to the first aspect or any one of the possible designs of the first aspect.
In a fourth aspect, the present invention provides a storage medium having stored thereon instructions for performing the article recommendation method as described in the first aspect or any one of the possible designs of the first aspect, when the instructions are run on a computer.
In a fifth aspect, the present invention provides a computer program product comprising instructions which, when run on a computer, cause the computer to perform the article recommendation method as described in the first aspect or any one of the possible designs of the first aspect.
The invention has the technical effects that:
(1) the invention provides an article recommendation scheme adopting a pentagonal average weighting algorithm, which comprises the steps of firstly obtaining an article hotspot index value of an article to be recommended, obtaining an article characteristic index value, a user recommendation refusing index value and a user reading index value of the article to be recommended on a target user, then calculating a comprehensive recommendation index value of the article to be recommended by adopting the pentagonal average weighting algorithm from five aspects of article hotspots, article characteristics, user recommendation refusing behaviors, user reading behaviors and the like, and finally deciding whether to push the article to be recommended according to a comparison result of the comprehensive recommendation index value and a threshold value, so that whether the target user is interested in predicting the article to be recommended can be predicted, the purpose of comprehensively expressing user behaviors and information attributes of the article can be realized, and the problem of insufficient independent images in a traditional recommendation system can be solved, the method is further beneficial for the recommender to suggest more interesting articles to the target user, improves the article pushing accuracy and is convenient for practical application and popularization;
(2) because the article recommendation method provided by the invention not only considers the characteristics of the articles and the users, but also considers whether the users start some article for reading, behaviors (such as recommending five articles, which are not read.
Drawings
In order to more clearly illustrate the embodiments of the present invention 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 of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a flow chart of an article recommendation method provided by the present invention.
Fig. 2 is an exemplary graph comparing experimental results of the article recommendation method, the existing hidden died allocation method, and the existing convolution matrix decomposition method provided by the present invention on the area index under the curve.
Fig. 3 is a schematic structural diagram of an article recommendation device provided in the present invention.
Fig. 4 is a schematic structural diagram of a computer device provided by the present invention.
Detailed Description
The invention is further described with reference to the following figures and specific embodiments. It should be noted that the description of the embodiments is provided to help understanding of the present invention, but the present invention is not limited thereto. Specific structural and functional details disclosed herein are merely representative of exemplary embodiments of the invention. This invention may, however, be embodied in many alternate forms and should not be construed as limited to the embodiments set forth herein.
It should be understood that, for the term "and/or" as may appear herein, it is merely an associative relationship that describes an associated object, meaning that three relationships may exist, e.g., a and/or B may mean: a exists alone, B exists alone, and A and B exist at the same time; for the term "/and" as may appear herein, which describes another associative object relationship, it means that two relationships may exist, e.g., a/and B, may mean: a exists independently, and A and B exist independently; in addition, for the character "/" that may appear herein, it generally means that the former and latter associated objects are in an "or" relationship.
As shown in fig. 1-2, the article recommendation method provided in the first aspect of the present embodiment may be executed by, but not limited to, a server held by an information service provider and used for responding to an access from a client, such as a web browser or APP (Application program, abbreviation), so as to help the information service provider to suggest more interesting articles to a target user, and ensure the accuracy of article pushing. The article recommendation method may include, but is not limited to, the following steps S1 to S3.
S1, obtaining an article hotspot index value of an article to be recommended, obtaining an article characteristic index value, a user recommendation refusing index value and a user reading index value of the article to be recommended on a target user, wherein the article hotspot index value is used for reflecting the similarity degree of the article to be recommended and a real-time hotspot news article, the article characteristic index value is used for reflecting the similarity degree of the article to be recommended and the prior reading article of the target user, the user characteristic index value is used for reflecting how much the similar users of the target user read the article to be recommended, the user rejections index value is used for reflecting the similarity degree of the article to be recommended and the prior rejections article of the target user, the user reading index value is used for reflecting the behavior depth of the target user when reading the previous similar articles of the articles to be recommended.
In the step S1, it is preferable to acquire the article hotspot index value of the article to be recommended, including but not limited to the following steps S111 to S114.
And S111, acquiring at least one hot news article in real time.
In the step S111, the top five news contents of the current hot headline may be screened out by the real-time news headline to serve as the at least one hot news article.
And S112, for each hot news article in the article to be recommended and the at least one hot news article, importing the corresponding article content into the trained neural network model, and obtaining the corresponding article characteristic value.
In step S112, the neural network model is a conventional model based on an existing artificial intelligence technology, and the trained neural network model can be obtained by an existing training method. In detail, the neural network model is preferably adopted as a pretraining model, BERT (bidirectional Encoder reproduction from transformations), which is proposed by Google AI research institute in 2018 and 10 months, and shows striking performance in machine reading understanding top level test SQuAD1.1, wherein the model surpasses human beings in all two metrics and becomes a milestone model achievement in NLP (Natural Language Processing) development history.
S113, calculating the similarity between the article to be recommended and each hot news article according to a cosine similarity method of the following formula:
wherein i represents a positive integer,
representing the similarity between the article to be recommended and the ith hot news article in the at least one hot news article, fPAn article feature value representing the article to be recommended,
an article feature value representing the ith hot news article.
And S114, according to the maximum similarity obtained by calculation and the preset corresponding relation between the multiple pairs of similarity intervals and the numerical values, taking the numerical value corresponding to the similarity interval where the maximum similarity exists as the hot point index value of the article.
In step S114, the preset corresponding relationship between the multiple pairs of similarity intervals and the numerical values may be, for example: for the similarity interval not greater than 0.5, the corresponding numerical value is preset to be 0; for the similarity interval greater than 0.5 and not greater than 0.7, the corresponding numerical value is preset to be 0.5; for a similarity interval greater than 0.7, the corresponding value is preset to 1. Thus, when the maximum similarity is 0.3, the value 0 can be used as the article hotspot index value; when the maximum similarity is 0.6, a value of 0.5 may be taken as the article hotspot index value; when the maximum similarity is 0.9, a value of 1 may be taken as the article hotspot index value.
In the step S1, it is preferable that the article feature index value of the article to be recommended on the target user is obtained, including but not limited to the following steps S121 to S124.
S121, acquiring all previous reading articles of the target user.
And S122, aiming at the article to be recommended and each of all the prior reading articles, importing the corresponding article content into the trained neural network model, and acquiring the corresponding article characteristic value.
In the step S122, the specific description of the neural network model may refer to the step S112, which is not described herein again.
S123, calculating the similarity between the article to be recommended and each of the prior reading articles according to a cosine similarity method of the following formula:
wherein j represents a positive integer,
representing the similarity of the article to be recommended and the jth prior reading article in all the prior reading articles, fPAn article feature value representing the article to be recommended,
an article feature value representing the jth prior-read article.
And S124, according to the maximum similarity obtained by calculation and the preset corresponding relation between the multiple pairs of similarity intervals and the numerical values, taking the numerical value corresponding to the similarity interval where the maximum similarity exists as the article characteristic index value.
In the step S124, for a detailed description of the preset corresponding relationship between the multiple pairs of similarity intervals and the numerical values, refer to the step S114, which is not described herein again.
In the step S1, it is preferable that the user characteristic index value of the article to be recommended on the target user is obtained, including but not limited to the following steps S131 to S135.
S131, acquiring the personal data of the target user and the personal data of all other users.
In the step S131, the other users are relative to the target user; the personal data can be uploaded by the user at the time of user registration, and specifically, but not limited to, the personal data can include information such as gender, age, school calendar, graduation professional and job position.
S132, aiming at the target user and each other user in all other users, importing the corresponding personal data into the trained neural network model, and obtaining the corresponding user characteristic value.
In the step S132, the detailed description of the neural network model can refer to the step S112, which is not described herein again.
S133, calculating the similarity between the target user and each other user according to a cosine similarity method of the following formula:
in the formula, k represents a positive integer,
representing the similarity of the target user to the kth other user among the all other users, fUA user characteristic value representing the target user,
and representing the user characteristic value of the kth other user.
And S134, screening out a plurality of other users with the former similarity according to the calculation result.
In the step S134, five other users with the top similarity may be screened out to serve as a similar user group most similar to the target user.
S135, counting the number of users reading the article to be recommended or the user number ratio among the other users, and taking the number of users or the user number ratio as the user characteristic index value.
In the step S1, it is preferable to obtain the user recommendation refusal index value of the article to be recommended on the target user, including but not limited to the following steps S141 to S144.
S141, acquiring all prior recommendation refusing articles of the target user, wherein the prior recommendation refusing articles refer to prior recommendation articles which are not read by the target user.
In the step S141, whether the previous recommended article is read by the target user may be determined based on a conventional data acquisition function added to the website code, for example, when a content loading progress acquired for a certain previous recommended article is always 0, which indicates that the previous recommended article is not read by the target user, the previous recommended article is the previous recommended refused article, otherwise, the previous recommended article is the previous read article.
And S142, for each article to be recommended and each article rejected in all the articles rejected in advance, importing the corresponding article content into the trained neural network model, and obtaining the corresponding article characteristic value.
In the step S142, the specific description of the neural network model may refer to the step S112, which is not described herein again.
S143, calculating the similarity between the article to be recommended and each of the prior recommendation refusing articles according to a cosine similarity method of the following formula:
wherein, x represents a positive integer,
representing a similarity of the article to be recommended and an xth preceding recommended article among the all preceding recommended articles, fPAn article feature value representing the article to be recommended,
an article characteristic value representing the xth prior rejector article.
And S144, according to the maximum similarity obtained by calculation and the preset corresponding relation between the multiple pairs of similarity intervals and the numerical values, taking the numerical value corresponding to the similarity interval where the maximum similarity exists as the user recommendation refusing index value.
In the step S144, for a detailed description of the preset corresponding relationship between the multiple pairs of similarity intervals and the numerical values, reference may be made to the step S114, which is not described herein again.
In the step S1, it is preferable that the user reading index value of the article to be recommended on the target user is obtained, including but not limited to the following steps S151 to S156.
S151, acquiring all previous reading articles of the target user.
S152, aiming at the article to be recommended and each of all the prior reading articles, importing the corresponding article content into the trained neural network model, and obtaining the corresponding article characteristic value.
In the step S152, the detailed description of the neural network model can refer to the step S112, which is not described herein again.
S153, calculating the similarity between the article to be recommended and each of the prior reading articles according to a cosine similarity method of the following formula:
wherein y represents a positive integer,
representing the similarity of the article to be recommended and the y-th prior reading article in all the prior reading articles, fPAn article feature value representing the article to be recommended,
an article feature value representing the y-th prior-read article.
S154, screening a plurality of prior reading articles with the former similarity according to the calculation result.
In the step S154, for example, five previous reading articles with the top similarity are screened out to be used as the previous similar article group most similar to the article to be recommended.
And S155, aiming at each prior reading article in the prior reading articles, determining a corresponding individual behavior depth value according to the corresponding reading behavior acquisition data.
In the step S155, the reading behavior collection data may be collected based on a conventional data collection function added to the website code, and includes, but is not limited to, a content loading progress, a reading time, a praise behavior record, a collection behavior record, a sharing behavior record, and/or a comment behavior record. Specifically, for each of the multiple previous reading articles, the corresponding individual behavior depth value is determined according to the corresponding reading behavior acquisition data, which includes, but is not limited to, the following steps S1551 to S1553.
S1551, initializing the fast reading behavior index value, the half reading behavior index value, the refined reading behavior index value, the like behavior index value, the collection behavior index value, the sharing behavior index value and/or the comment behavior index value to zero.
S1552, according to reading behavior acquisition data corresponding to a previous reading article, updating the fast reading behavior index value, the semi-reading behavior index value, the refined reading behavior index value, the like behavior index value, the collected behavior index value, the shared behavior index value and/or the comment behavior index value in the following modes: when the reading behavior collection data indicates that the previous reading article is quickly read by the target user (which may be specifically determined based on the content loading progress and the reading time, for example, when the content loading progress is 100% and the reading time is lower than a preset time threshold), updating the quick reading behavior index value to a first numerical value (for example, 1) greater than zero; when the reading behavior collection data indicates that the previous reading article is not read completely by the target user and quits reading (which may be determined based on the content loading progress, for example, when the content loading progress is 60%), updating the half-reading behavior index value to a second value (for example, -1) smaller than zero; when the reading behavior collection data indicates that the previous reading article is perused by the target user (which may be specifically determined based on the content loading progress and the reading time, for example, when the content loading progress is 100% and the reading time is higher than a preset time threshold), updating the perusing behavior index value to be a third value (for example, 1) greater than zero; when the reading behavior collection data indicates that the previous reading article is praised by the target user (specifically, the praise behavior collection data can be judged based on a praise behavior record), updating the praise behavior index value to a fourth numerical value (for example, 1) larger than zero; when the reading behavior collection data indicates that the previous reading article is collected by the target user (specifically, the previous reading article can be judged based on a collection behavior record), updating the collection behavior index value to be a fifth numerical value (for example, 1) larger than zero; when the reading behavior collection data indicates that the previous reading article is shared by the target user (which may be specifically determined based on the sharing behavior record), updating the sharing behavior index value to a sixth numerical value (for example, 1) greater than zero; when the reading behavior collection data indicates that the previous reading article is reviewed by the target user (which may be specifically determined based on review behavior records), the review behavior index value is updated to a seventh numerical value (e.g., 1) greater than zero.
S1553, taking the average value or the accumulated value of the quick reading behavior index value, the semi-reading behavior index value, the refined reading behavior index value, the like behavior index value, the collection behavior index value, the sharing behavior index value and/or the comment behavior index value as the individual behavior depth value corresponding to the prior reading article.
S156, calculating according to the following formula to obtain the user reading index value A5：
Wherein M represents the number of articles of the plurality of prior reading articles, z represents a positive integer,
a predetermined weight coefficient representing a z-th prior-read article among the plurality of prior-read articles, azAnd the individual behavior depth value of the z th prior reading article in the plurality of prior reading articles is represented, and the preset weight coefficient of the prior reading article is positively correlated with the corresponding similarity or negatively correlated with the corresponding sequence number which is obtained by sequencing from high to low according to the similarity.
In the step S156, for example, when M is 5, for the top five previous reading articles in the sequence, the corresponding preset weight coefficients may be preset to 0.5, 0.4, 0.3, 0.25, and 0.2 in sequence according to the sequence from high to low corresponding to the similarity, that is, the weight coefficients are negatively correlated to the corresponding sequence numbers and are obtained by the sequence from high to low corresponding to the similarity.
S2, according to the article hotspot index value, the article characteristic index value, the user recommendation refusing index value and the user reading index value, calculating to obtain a comprehensive recommendation index value A of the article to be recommended according to the following formulaP：
AP＝w1*A1+w2*A2+w3*A3+w4*A4+w5*A5
In the formula, w1Representing a preset weight coefficient, A, corresponding to the article hotspot index value1Representing the article hotspot index value, w2Representing a preset weight coefficient, A, corresponding to the article feature index value2Representing the value of the article characteristic index, w3Representing a preset weight coefficient corresponding to said user characteristic index value, A3Representing said user characteristic index value, w4Representing a preset weight coefficient corresponding to the user recommendation rejection index value, A4Represents the user recommendation refusal index value w5Representing a preset weighting factor, A, corresponding to said user reading index value5And representing the user reading index value.
In step S2, the calculation method is a pentagonal average weighting algorithm, that is, a comprehensive recommendation index value of the article to be recommended is calculated from five major angles, that is, an article hotspot, an article characteristic, a user recommendation refusing behavior, a user reading behavior, and the like, so as to predict whether the target user is interested in the article to be recommended, and achieve the purpose of comprehensively expressing the user behavior and the information attribute of the article. In addition, the maximum possible values of the article hotspot index value, the article characteristic index value, the user referral rejection index value and the user reading index value can be respectively limited to 1 (for example, through normalization processing), and their corresponding preset weighting coefficients can be 0.3, 0.2, -0.3 (i.e., one recommendation deduction item in terms of user referral behavior) and 0.3 in sequence.
S3, when the comprehensive recommended index value A is judgedPAnd when the recommendation index is larger than or equal to a preset recommendation index threshold value, pushing the article to be recommended to the target user.
In the step S3, when the maximum possible values of the article hotspot index value, the article feature index value, the user recommendation refusal index value and the user reading index value are respectively limited to 1, the preset recommendation index threshold may be 0.5, for example.
Based on the foregoing steps S1 to S3, the applicant further specifically develops an article recommendation system website for performing experiments, and writes corresponding data acquisition embedded point codes in the web pages of the website to acquire the following types of data: the data may be obtained by recording the Decision Process of the user by using a Markov Decision Process (MDP), which is a mathematical model of sequential Decision and is used to simulate a randomness policy and a reward that can be realized by an agent in an environment where a system state has a Markov property. In addition, in the experimental stage, the applicant collects 500 data samples of 10 users, and compared with the mainstream recommendation method: hidden Dirichlet Allocation (LDA) and convolution Matrix decomposition (ConvmF) were compared, and the comparison results are shown in Table 1 below:
TABLE 1 comparison of three metrics for the recommended method
As can be seen from table 1, the article recommendation method of the present invention is based on three metrics: the Area Under the Curve (AUC), Mean Reciprocal Rank (MRR) and K-normalized discounted cumulative gain (K is 5, NDCG @ K) are all better. As also shown in fig. 2, it shows the case that the recommendation accuracy of the three recommendation methods increases with the increase of the number of clicks of the user in the experimental process: the three recommendation methods become more accurate along with the participation of the user (clicking the recommended articles), and the recommendation method of the invention exceeds the other two methods after the number of clicks of the user is increased.
Thus, through the article recommendation method described in detail in the above steps S1 to S3, an article recommendation scheme using a pentagonal average weighting algorithm can be provided, that is, an article hotspot index value of an article to be recommended is obtained, an article characteristic index value, a user recommendation rejection index value, and a user reading index value of the article to be recommended on a target user are obtained, then a pentagonal average weighting algorithm is used to calculate a comprehensive recommendation index value of the article to be recommended from five points of the article hotspot, the article characteristic, the user recommendation rejection behavior, the user reading behavior, and the like, and finally whether the article to be recommended is to be pushed is decided according to a comparison result of the comprehensive recommendation index value and a threshold value, so that whether the target user is interested in predicting the article to be recommended can be predicted, and the purpose of comprehensively expressing user behavior and information attributes of the article can be realized, the problem of insufficient independent images in a traditional recommendation system is solved, and then a recommender is helped to suggest more interesting articles to a target user, the article pushing accuracy is improved, and the practical application and popularization are facilitated. In addition, because the article recommendation method of the invention not only considers the characteristics of the articles and the users, but also considers whether the users start some article for reading, and behaviors (such as recommending five articles, which are not read.
As shown in fig. 3, a second aspect of this embodiment provides a virtual device for implementing the method of the first aspect or any one of the possible designs of the first aspect, including an index obtaining module, an index synthesizing module, and an article pushing module, which are sequentially connected in a communication manner;
the index acquisition module is used for acquiring article hotspot index values of articles to be recommended, acquiring article characteristic index values, user recommendation refusing index values and user reading index values of the articles to be recommended on a target user, wherein the article hotspot index value is used for reflecting the similarity degree of the article to be recommended and a real-time hotspot news article, the article characteristic index value is used for reflecting the similarity degree of the article to be recommended and the prior reading article of the target user, the user characteristic index value is used for reflecting how much the similar users of the target user read the article to be recommended, the user rejections index value is used for reflecting the similarity degree of the article to be recommended and the prior rejections article of the target user, the user reading index value is used for reflecting the behavior depth of the target user when reading the previous similar articles of the articles to be recommended;
the index integration module is used for calculating an integrated recommendation index value A of the article to be recommended according to the article hotspot index value, the article characteristic index value, the user recommendation refusing index value and the user reading index value and according to the following formulaP：
AP＝w1*A1+w2*A2+w3*A3+w4*A4+w5*A5
In the formula, w1Representing a preset weight coefficient, A, corresponding to the article hotspot index value1Representing the article hotspot index value, w2Representing a preset weight coefficient, A, corresponding to the article feature index value2Representing the value of the article characteristic index, w3Representing a preset weight coefficient corresponding to said user characteristic index value, A3Representing said user characteristic index value, w4Representing a preset weight coefficient corresponding to the user recommendation rejection index value, A4Represents the user recommendation refusal index value w5Representing a preset weighting factor, A, corresponding to said user reading index value5Representing the user reading index value;
the article pushing module is used for judging the comprehensive recommended index value APWhen the recommendation index is larger than or equal to a preset recommendation index threshold value, pushing the target userThe article to be recommended.
For the working process, working details and technical effects of the foregoing apparatus provided in the second aspect of this embodiment, reference may be made to the method described in the first aspect or any one of the possible designs of the first aspect, which is not described herein again.
As shown in fig. 4, a third aspect of this embodiment provides a computer device for executing the method according to any one of the first aspect or the possible designs of the first aspect, where the computer device includes a memory and a processor, the memory is used for storing a computer program, and the processor is used for reading the computer program and executing the article recommendation method according to any one of the first aspect or the possible designs of the first aspect. For example, the Memory may include, but is not limited to, a Random-Access Memory (RAM), a Read-Only Memory (ROM), a Flash Memory (Flash Memory), a First-in First-out (FIFO), and/or a First-in Last-out (FILO), and the like; the processor may not be limited to the use of a microprocessor of the model number STM32F105 family. In addition, the computer device may also include, but is not limited to, a power module, a display screen, and other necessary components.
For the working process, working details, and technical effects of the foregoing computer device provided in the third aspect of this embodiment, reference may be made to the method in the first aspect or any one of the possible designs in the first aspect, which is not described herein again.
A fourth aspect of the present embodiment provides a storage medium storing instructions of the method according to any one of the first aspect or the possible designs of the first aspect, that is, the storage medium has instructions stored thereon, and when the instructions are executed on a computer, the article recommendation method according to any one of the first aspect or the possible designs of the first aspect is executed. The storage medium refers to a carrier for storing data, and may include, but is not limited to, a computer-readable storage medium such as a floppy disk, an optical disk, a hard disk, a flash Memory, a flash disk and/or a Memory Stick (Memory Stick), and the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.
For the working process, the working details, and the technical effects of the foregoing storage medium provided in the fourth aspect of this embodiment, reference may be made to the method in the first aspect or any one of the possible designs in the first aspect, which is not described herein again.
A fifth aspect of the present embodiments provides a computer program product comprising instructions which, when run on a computer, cause the computer to perform the article recommendation method as described in the first aspect or any one of the possible designs of the first aspect. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable devices.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: modifications may be made to the embodiments described above, or equivalents may be substituted for some of the features described. And such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Finally, it should be noted that the present invention is not limited to the above alternative embodiments, and that various other forms of products can be obtained by anyone in light of the present invention. The above detailed description should not be taken as limiting the scope of the invention, which is defined in the claims, and which the description is intended to be interpreted accordingly.
Claims (10)
1. An article recommendation method, comprising:
obtaining an article hotspot index value of an article to be recommended, obtaining an article characteristic index value, a user recommendation refusing index value and a user reading index value of the article to be recommended on a target user, wherein the article hotspot index value is used for reflecting the similarity degree of the article to be recommended and a real-time hotspot news article, the article characteristic index value is used for reflecting the similarity degree of the article to be recommended and the prior reading article of the target user, the user characteristic index value is used for reflecting how much the similar users of the target user read the article to be recommended, the user rejections index value is used for reflecting the similarity degree of the article to be recommended and the prior rejections article of the target user, the user reading index value is used for reflecting the behavior depth of the target user when reading the previous similar articles of the articles to be recommended;
according to the article hotspot index value, the article characteristic index value, the user recommendation refusing index value and the user reading index value, calculating a comprehensive recommendation index value A of the article to be recommended according to the following formulaP：
AP＝w1*A1+w2*A2+w3*A3+w4*A4+w5*A5
In the formula, w1Representing a preset weight coefficient, A, corresponding to the article hotspot index value1Representing the article hotspot index value, w2Representing a preset weight coefficient, A, corresponding to the article feature index value2Representing the value of the article characteristic index, w3Representing a preset weight coefficient corresponding to said user characteristic index value, A3Representing said user characteristic index value, w4Representing a preset weight coefficient corresponding to the user recommendation rejection index value, A4Represents the user recommendation refusal index value w5Representing a preset weighting factor, A, corresponding to said user reading index value5Representing the user reading index value;
when the comprehensive recommended index value A is judgedPAnd when the recommendation index is larger than or equal to a preset recommendation index threshold value, pushing the article to be recommended to the target user.
2. The article recommendation method of claim 1, wherein obtaining the article hotspot index value of the article to be recommended comprises:
acquiring at least one hot news article in real time;
for each hot news article in the article to be recommended and the at least one hot news article, importing the corresponding article content into a trained neural network model, and obtaining a corresponding article characteristic value;
and calculating the similarity between the article to be recommended and each hot news article according to a cosine similarity method of the following formula:
wherein i represents a positive integer,
representing the similarity between the article to be recommended and the ith hot news article in the at least one hot news article, fPAn article feature value representing the article to be recommended,
an article feature value representing the ith hot news article;
and according to the maximum similarity obtained by calculation and the preset corresponding relation between the multiple pairs of similarity intervals and the numerical value, taking the numerical value corresponding to the similarity interval where the maximum similarity exists as the article hotspot index value.
3. The article recommendation method of claim 1, wherein obtaining the article feature index value of the article to be recommended on the target user comprises:
acquiring all prior reading articles of the target user;
for the article to be recommended and each of the all the previously read articles, importing the corresponding article content into a trained neural network model, and acquiring a corresponding article characteristic value;
and calculating the similarity between the article to be recommended and each of the prior reading articles according to a cosine similarity method of the following formula:
wherein j represents a positive integer,
representing the similarity of the article to be recommended and the jth prior reading article in all the prior reading articles, fPAn article feature value representing the article to be recommended,
an article feature value representing the jth prior-read article;
and taking the numerical value corresponding to the similarity interval with the maximum similarity as the article characteristic index value according to the maximum similarity obtained by calculation and the preset corresponding relation between the multiple pairs of similarity intervals and the numerical value.
4. The article recommendation method of claim 1, wherein obtaining the user characteristic index value of the article to be recommended on the target user comprises:
acquiring personal data of the target user and personal data of all other users;
aiming at the target user and each other user in all other users, importing the corresponding personal data into the trained neural network model to obtain the corresponding user characteristic value;
and calculating the similarity between the target user and each other user according to a cosine similarity method of the following formula:
in the formula, k tableA positive integer is shown as a whole number,
representing the similarity of the target user to the kth other user among the all other users, fUA user characteristic value representing the target user,
a user characteristic value representing the kth other user;
screening out a plurality of other users with the former similarity according to the calculation result;
and counting the number of users reading the article to be recommended or the user number ratio among the other users, and taking the number of users or the user number ratio as the user characteristic index value.
5. The article recommendation method of claim 1, wherein obtaining the user recommendation rejection index value of the article to be recommended on the target user comprises:
acquiring all prior recommendation refusing articles of the target user, wherein the prior recommendation refusing articles refer to prior recommendation articles which are not read by the target user;
for the article to be recommended and each of the articles rejected earlier, importing the corresponding article content into a trained neural network model to obtain a corresponding article characteristic value;
calculating the similarity between the article to be recommended and each of the prior recommended refusing articles according to a cosine similarity method of the following formula:
wherein, x represents a positive integer,
representing the article to be recommendedSimilarity to the xth of said all previous rejections articles, fPAn article feature value representing the article to be recommended,
an article characteristic value representing the xth prior rejector article;
and according to the maximum similarity obtained by calculation and the preset corresponding relation between the multiple pairs of similarity intervals and the numerical value, taking the numerical value corresponding to the similarity interval with the maximum similarity as the user recommendation refusing index value.
6. The article recommendation method of claim 1, wherein obtaining the user reading index value of the article to be recommended on the target user comprises:
acquiring all prior reading articles of the target user;
for the article to be recommended and each of the all the previously read articles, importing the corresponding article content into a trained neural network model, and acquiring a corresponding article characteristic value;
and calculating the similarity between the article to be recommended and each of the prior reading articles according to a cosine similarity method of the following formula:
wherein y represents a positive integer,
representing the similarity of the article to be recommended and the y-th prior reading article in all the prior reading articles, fPAn article feature value representing the article to be recommended,
text representing the y-th prior-reading articleA chapter feature value;
screening a plurality of prior reading articles with the former similarity according to the calculation result;
for each of the plurality of prior reading articles, determining a corresponding individual behavior depth value according to corresponding reading behavior acquisition data;
calculating the reading index value A of the user according to the following formula5：
Wherein M represents the number of articles of the plurality of prior reading articles, z represents a positive integer,
a predetermined weight coefficient representing a z-th prior-read article among the plurality of prior-read articles, azAnd the individual behavior depth value of the z th prior reading article in the plurality of prior reading articles is represented, and the preset weight coefficient of the prior reading article is positively correlated with the corresponding similarity or negatively correlated with the corresponding sequence number which is obtained by sequencing from high to low according to the similarity.
7. An article recommendation method according to claim 6, wherein determining, for each of the plurality of prior reading articles, a corresponding individual behavior depth value from the corresponding reading behavior collection data comprises:
initializing the fast reading behavior index value, the semi-reading behavior index value, the refined reading behavior index value, the like behavior index value, the collection behavior index value, the sharing behavior index value and/or the comment behavior index value to zero;
according to reading behavior collection data corresponding to a previous reading article, updating the fast reading behavior index value, the semi-reading behavior index value, the refined reading behavior index value, the like behavior index value, the collection behavior index value, the sharing behavior index value and/or the comment behavior index value according to the following modes:
when the reading behavior acquisition data shows that the prior reading article is quickly read by the target user, updating the quick reading behavior index value to be a first numerical value larger than zero;
when the reading behavior acquisition data shows that the prior reading article is not read completely by the target user and then quits reading, updating the semi-reading behavior index value to be a second value smaller than zero;
when the reading behavior acquisition data shows that the prior reading article is accurately read by the target user, updating the accurate reading behavior index value to be a third numerical value larger than zero;
when the reading behavior collection data shows that the prior reading article is praised by the target user, updating the praise behavior index value to be a fourth numerical value larger than zero;
when the reading behavior acquisition data show that the prior reading article is collected by the target user, updating the collection behavior index value to be a fifth numerical value larger than zero;
when the reading behavior acquisition data shows that the prior reading article is shared by the target user, updating the sharing behavior index value to be a sixth numerical value larger than zero;
when the reading behavior collection data show that the prior reading article is commented by the target user, updating the comment behavior index value to be a seventh numerical value larger than zero;
taking the average value or the accumulated value of the fast reading behavior index value, the half reading behavior index value, the refined reading behavior index value, the like behavior index value, the collection behavior index value, the sharing behavior index value and/or the comment behavior index value as the individual behavior depth value corresponding to the prior reading article.
8. An article recommending device is characterized by comprising an index obtaining module, an index integrating module and an article pushing module which are sequentially in communication connection;
the index acquisition module is used for acquiring article hotspot index values of articles to be recommended, acquiring article characteristic index values, user recommendation refusing index values and user reading index values of the articles to be recommended on a target user, wherein the article hotspot index value is used for reflecting the similarity degree of the article to be recommended and a real-time hotspot news article, the article characteristic index value is used for reflecting the similarity degree of the article to be recommended and the prior reading article of the target user, the user characteristic index value is used for reflecting how much the similar users of the target user read the article to be recommended, the user rejections index value is used for reflecting the similarity degree of the article to be recommended and the prior rejections article of the target user, the user reading index value is used for reflecting the behavior depth of the target user when reading the previous similar articles of the articles to be recommended;
the index integration module is used for calculating an integrated recommendation index value A of the article to be recommended according to the article hotspot index value, the article characteristic index value, the user recommendation refusing index value and the user reading index value and according to the following formulaP：
AP＝w1*A1+w2*A2+w3*A3+w4*A4+w5*A5
In the formula, w1Representing a preset weight coefficient, A, corresponding to the article hotspot index value1Representing the article hotspot index value, w2Representing a preset weight coefficient, A, corresponding to the article feature index value2Representing the value of the article characteristic index, w3Representing a preset weight coefficient corresponding to said user characteristic index value, A3Representing said user characteristic index value, w4Representing a preset weight coefficient corresponding to the user recommendation rejection index value, A4Represents the user recommendation refusal index value w5Representing a preset weighting factor, A, corresponding to said user reading index value5Representing the user reading index value;
the article pushing moduleA block for determining the integrated recommended index value A when it is determinedPAnd when the recommendation index is larger than or equal to a preset recommendation index threshold value, pushing the article to be recommended to the target user.
9. A computer device comprising a memory and a processor communicatively coupled, wherein the memory is configured to store a computer program, and the processor is configured to read the computer program and execute the article recommendation method of any one of claims 1-7.
10. A storage medium having stored thereon instructions for performing the article recommendation method of any one of claims 1-7 when the instructions are run on a computer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110545897.6A CN113239182A (en) | 2021-05-19 | 2021-05-19 | Article recommendation method and device, computer equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110545897.6A CN113239182A (en) | 2021-05-19 | 2021-05-19 | Article recommendation method and device, computer equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113239182A true CN113239182A (en) | 2021-08-10 |
Family
ID=77137662
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110545897.6A Pending CN113239182A (en) | 2021-05-19 | 2021-05-19 | Article recommendation method and device, computer equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113239182A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113672718A (en) * | 2021-09-02 | 2021-11-19 | 杭州一知智能科技有限公司 | Dialog intention recognition method and system based on feature matching and field self-adaption |
| CN114491295A (en) * | 2022-04-14 | 2022-05-13 | 北京创新乐知网络技术有限公司 | Android-based network community resource recommendation method and device |
| CN116304128A (en) * | 2023-03-01 | 2023-06-23 | 广西泛华于成信息科技有限公司 | Multimedia information recommendation system based on big data |
-
2021
- 2021-05-19 CN CN202110545897.6A patent/CN113239182A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113672718A (en) * | 2021-09-02 | 2021-11-19 | 杭州一知智能科技有限公司 | Dialog intention recognition method and system based on feature matching and field self-adaption |
| CN113672718B (en) * | 2021-09-02 | 2024-04-05 | 杭州一知智能科技有限公司 | Dialogue intention recognition method and system based on feature matching and field self-adaption |
| CN114491295A (en) * | 2022-04-14 | 2022-05-13 | 北京创新乐知网络技术有限公司 | Android-based network community resource recommendation method and device |
| CN114491295B (en) * | 2022-04-14 | 2022-06-24 | 北京创新乐知网络技术有限公司 | Android-based network community resource recommendation method and device |
| CN116304128A (en) * | 2023-03-01 | 2023-06-23 | 广西泛华于成信息科技有限公司 | Multimedia information recommendation system based on big data |
| CN116304128B (en) * | 2023-03-01 | 2023-12-15 | 微众梦想科技(北京)有限公司 | Multimedia information recommendation system based on big data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Karimi et al. | News recommender systems–Survey and roads ahead | |
| CN110321422B (en) | Method for training model on line, pushing method, device and equipment | |
| CN108520076B (en) | Electronic book recommendation method, electronic device and computer storage medium | |
| TWI636416B (en) | Method and system for multi-phase ranking for content personalization | |
| US11698939B2 (en) | Prioritizing items from different categories in a news stream | |
| CN113239182A (en) | Article recommendation method and device, computer equipment and storage medium | |
| CN110598086B (en) | Article recommendation method and device, computer equipment and storage medium | |
| WO2015192667A1 (en) | Advertisement recommending method and advertisement recommending server | |
| CN110532479A (en) | A kind of information recommendation method, device and equipment | |
| CN110704674A (en) | Video playing integrity prediction method and device | |
| CN106599047B (en) | Information pushing method and device | |
| CN111159341B (en) | Information recommendation method and device based on user investment and financial management preference | |
| CN111461841A (en) | Article recommendation method, device, server and storage medium | |
| CN111400546B (en) | Video recall method and video recommendation method and device | |
| Wang et al. | Modeling uncertainty to improve personalized recommendations via Bayesian deep learning | |
| CN111597446B (en) | Content pushing method and device based on artificial intelligence, server and storage medium | |
| CN110598126B (en) | Cross-social network user identity recognition method based on behavior habits | |
| CN114265953A (en) | Short video recommendation method, system, device and medium based on label | |
| CN112565902B (en) | Video recommendation method and device and electronic equipment | |
| CN114581165A (en) | Product recommendation method, device, computer storage medium and system | |
| CN110569447B (en) | Network resource recommendation method and device and storage medium | |
| CN113836388A (en) | Information recommendation method and device, server and storage medium | |
| CN110727867A (en) | Semantic entity recommendation method based on fuzzy mechanism | |
| Bunga et al. | From implicit preferences to ratings: video games recommendation based on collaborative filtering | |
| CN110633408A (en) | Recommendation method and system for intelligent business information |
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 |