CN113065062A - News recommendation method and system based on user reading time behavior - Google Patents
News recommendation method and system based on user reading time behavior Download PDFInfo
- Publication number
- CN113065062A CN113065062A CN202110258628.1A CN202110258628A CN113065062A CN 113065062 A CN113065062 A CN 113065062A CN 202110258628 A CN202110258628 A CN 202110258628A CN 113065062 A CN113065062 A CN 113065062A
- Authority
- CN
- China
- Prior art keywords
- news
- user
- reading
- time
- reading speed
- 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
- 238000000034 method Methods 0.000 title claims abstract description 28
- 230000006399 behavior Effects 0.000 claims abstract description 22
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000013145 classification model Methods 0.000 claims description 3
- 230000004931 aggregating effect Effects 0.000 claims description 2
- 238000013527 convolutional neural network Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 239000000126 substance Substances 0.000 claims description 2
- 238000001914 filtration Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a news recommendation method and a system based on user reading time behaviors. For the news which is exposed and not read, the method adopts a spy algorithm to distinguish the news which is really uninteresting to the user from the news which is potentially interesting, so that the quality and the diversity of the recommended news are ensured. The invention solves the problems of timeliness, title misleading and recommendation content unicity of the personalized news recommendation system. Abstracting the reading time behavior of each user into personalized reading speed, introducing an attention mechanism, and measuring different influence degrees of historical click news of the user on candidate news; for news which is not clicked by the exposed user, a spy algorithm is introduced to distinguish the news in which the user really does not interest and the news in which the user has potential interest, so that the quality and the diversity of the recommended news are ensured.
Description
Technical Field
The invention relates to the technical field of personalized news recommendation, in particular to a recommendation system for personalized news recommendation based on user reading time.
Background
With the continuous development of information technology and the internet, people gradually move from the times of lacking information to the times of information overload. In this age, both information consumers and information producers face significant challenges. For information consumers, it is very difficult to obtain valuable information from a large amount of information; it is also very difficult for information producers to make their own information stand out and to present the information to information consumers. In order to solve the problem of information overload, two tools of a search engine and a recommendation system are brought forward. Nowadays, recommendation systems are gradually applied to various fields of the internet, which mainly analyze interests and hobbies of users through historical behavior records of the users to recommend the users, including e-commerce recommendation, personalized advertisement recommendation, news recommendation and the like, and are widely applied to products such as Taobao, today's top-of-the-day, trembled short videos and the like.
When the recommended data volume and the users grow in a large scale, the complexity of the recommendation system can also grow synchronously. The current mainstream recommendation algorithms include collaborative filtering recommendation algorithms, content-based recommendation algorithms and the like. The collaborative filtering recommendation method mainly comprises two methods: item-based collaborative filtering and user-based collaborative filtering. The basic idea of collaborative filtering based on articles is to find the similarity between items based on the evaluation of users on the items, and then recommend similar articles to the users according to the historical preference information of the users. The basic idea of the user-based collaborative filtering is to find neighbor users with similar interests as the target user and recommend items which are interested by the neighbor users and are not acted by the target user to the target user. The collaborative filtering algorithm does not need to obtain characteristic data of articles or users in advance, only depends on historical behaviors of the users to recommend the users, but for new users, the problem of cold start caused by the fact that no historical data exists can exist, and the recommendation effect can be influenced to a certain extent. The recommendation algorithm based on the content directly recommends the articles similar to the content in which the user is interested, for example, the user likes sports, the system recommends sports equipment and the like, historical behavior data of the user is not needed in the process, and therefore the problem of cold start of recommendation can be well solved.
The personalized recommendation system mainly researching news is different from other recommendation systems such as e-commerce recommendation, advertisement recommendation and the like, and the news recommendation has the following challenges: (1) the news is highly sensitive to time, has strong timeliness and is updated quickly. Studies have shown that approximately 85% of news articles will no longer be displayed on the news home page two days after first appearance, and thus the time of news production is an important factor for news recommendation; (2) the headlines of the news are highly concentrated, while the text is of a longer length. Statistical results show that the length of a news title is generally 5 to 15 words, the space of contents is generally more than 200 words, the prediction of the interest degree of a user on the news text through a highly condensed news title is inaccurate in some cases, and especially for some news with misleading titles, such as a common title party and the like, the user may feel disappointed in the contents after clicking the news; (3) people only are interested in a plurality of specific news categories in news reading, the conventional recommendation system analyzes and integrates the use data or browsing trace information of a user at a news client to push the news meeting the information requirements of the user, and has a certain effect.
Disclosure of Invention
The invention mainly aims to solve the problems in the conventional news recommending system to a certain extent, and provides a personalized news recommending method based on user reading time behaviors. According to the method, the time information of the user reading news is analyzed, so that an 'personalized reading speed' is generated for each user, and then recommendation is made by combining the timeliness of candidate news. For the news which is exposed and not read, the method adopts a spy algorithm to distinguish the news in which the user really does not interest and the news with potential interest, thereby ensuring the quality and diversification of the recommended news. The method and the system are used for solving the problems of timeliness, title misleading, recommendation content unicity and the like of the personalized news recommendation system.
The innovation points of the invention are as follows:
(1) the reading time behavior of each user is abstracted into 'personalized reading speed', and an attention mechanism is introduced according to the 'personalized reading speed', so that different influence degrees of historical click news of the user on candidate news are measured.
(2) For the news which is not clicked by the exposed user, a 'spy algorithm' is introduced to distinguish the news which is not really interested by the user from the news which is potentially interested by the user, so that the quality and the diversity of the recommended news can be ensured.
In order to achieve the above object, the present invention provides a news recommendation method based on user reading time behavior,
the method comprises the following specific steps:
(1) extracting various information of news read by a user, such as news titles, reading duration, news release time and the like, from a behavior log of the user;
(2) for each extracted news, constructing an individualized representation mode of the news by adopting a convolutional neural network;
(3) because the interest degree of each browsed news is different, different news is endowed with different weights through an attention mechanism;
(4) weighting and aggregating all news read by a user in a time range, wherein the weighted and aggregated news can be used for representing the interest direction of the user and is used as a characteristic representation of the user;
(5) for news which is exposed but not clicked by a user, a spy algorithm is adopted to find news which is potentially interesting to the user, and the news and newly generated news which is not recommended are added into candidate news to be recommended together for processing;
(6) and calculating the similarity between the candidate news to be recommended and the user characteristics, and recommending the news with higher similarity.
The recommendation method mainly measures the user by reading time information of each news of the userThe interest level of the news mainly comprises three aspects: the first is the news release time, and as news has strong timeliness, news with closer release time is more suitable for being recommended to users in general; secondly, the time length consumed by the user in the process of reading a certain news can be accurately reflected by the index, if the user clicks the news by being tempted by the title, the user finds that the news is uninteresting after browsing the content, the reading time is generally short, which is common in some 'headline party' news, the longer the reading time is, the higher the accuracy of news recommendation is explained, but the reading time duration is also influenced by the news space, so the reading time duration of the user is measured by adopting the reading time of the unit word number, the influence of the news space on the reading time duration is eliminated, however, the reading speed of each user is different, the same news is the same, the time required by different users for reading is different, and therefore, the invention provides a concept of 'personalized reading speed', firstly, through the historical reading behavior of a user, the reading word number of the user reading all news in unit time is calculated as the average reading speed v, and for a specific news, the reading word number of the user in unit time is calculated as the reading speed vtiBy vtiThe ratio of the average reading speed v to the reading speed v of the user can be well described for the specific news, when the ratio is larger than 1, the reading speed of the user is higher than the average reading speed and can be skipped, the user has common interest in the specific news, when the ratio is smaller than 1, the reading speed of the user is lower, the user has certain interest in the specific news, and the reading is in detail reading, so that the reading speed v of the user on the specific news can be well described by the method and the device, and the reading speed v of the user on the specific news is higher than the average reading speed vtiThe ratio of the average reading speed v to the average reading speed v is defined as the personalized reading speed, and the index can measure the interest degree of the user in the news; the third part is news with the reading time of 0, i.e. news recommended to the user and not clicked by the user, and for the part of news, it is not good to directly classify the news into news which is not interesting to the user because the reasons why the user does not click are various, for example, the news displayed on the page is too muchThe user may not have time to read all. On the other hand, the part of news is the news which is considered to be liked by the user and is selected by the recommendation system through calculation, if the news is directly listed as a negative sample, similar news is not recommended any more, contradictions can be caused, in the subsequent recommendation, the system is more prone to recommending news with higher hot spots, and the recommendation is not performed on news with relatively fewer clicks, so that the problem that the recommendation system does not meet the recommendation diversity is caused. In order to solve the contradiction, the invention provides a spy algorithm, which specifically comprises the steps of taking news clicked and browsed by a user as a positive sample set P, randomly selecting a part of subsets S from the positive sample set P, adding S into an unchecked news set to serve as a negative sample set N, then training an SVN classification model, scoring the samples in the set N by using a trained classifier, calculating an average score of the spy set S after scoring, dividing the samples with the score lower than the score in the negative sample set N into negative samples, and not dividing the samples with the score higher than the score into the negative samples, so that the news which is really uninterested by the user can be effectively distinguished.
The method mainly solves the problems of two aspects, the first aspect is to filter out interest embedding deviation caused by error clicking of a user through indexes such as personalized reading speed and the like, and endow the perusal news with lower weight for obviously higher reading speed so as to correct the interest model of the user; the second aspect is that for news exposed by the recommendation system but not clicked by the user, the classification of the news is reclassified by adopting a spy algorithm instead of uniformly classifying the news into negative samples.
Drawings
Fig. 1 is a schematic flow chart of a news recommendation system based on user reading time behavior according to the present invention.
Fig. 2 is a diagram illustrating different reading time behaviors of a user.
Detailed Description
The invention relates to a recommendation system for personalized news recommendation based on user reading time. In order to show the technical solution of the present invention more clearly, the technical solution of the present invention is further described in detail below with reference to the accompanying drawings.
As shown in fig. 1, the news recommendation method includes the steps of:
in the step (1), various information of news read by the user is extracted from the behavior log of the user, and the click history of the user i is obtained. Each log mainly comprises information such as a time stamp, a user ID, a news title and the like, and the click history of the user is represented asWhereinIs the jth news headline clicked by user i, N is the total number of news clicked by user i, t ═ w1,w2,w3,......]Title, w, representing newsiThis step initially extracts the news information for a word in the news headline and performs corresponding preprocessing.
In the step (2), the news headline t in the step (1) is converted into an embedded matrix W through a word vector model, and the sub-matrix W of the W is subjected toi,i+l-1Performing convolution operation to obtain characteristicsAfter performing convolution operation on each sub-matrix, a feature map can be obtained:
wherein h is a convolution kernel, l is the size of the convolution kernel, and f is a nonlinear function, a plurality of features can be obtained by using a plurality of convolution kernels with different sizes, and finally the features are connected in series to form a final news representation vector:
e(t)=[Ch1 Ch2......Chm] (2)
in step (3), the user pairs are differentThe interest degree of the news is different, the weight of the different news occupying the user preference needs to be calculated, and the method uses an attention mechanism to calculate. Representing the historical news in the step (2) into a vectorAs input, with the current candidate news tjRepresents the vector e (t)j) Calculating normalized influence weight:
the index can describe the influence degree of the historical news read by the user on the news to be recommended.
In the step (4), in order to express the interest degree of the user in different news, introducing personalized reading speed:
wherein the content of the first and second substances,indicating the number of words read by the user i per unit time when reading the j news, the index describing the reading speed of the user i when reading the j news,the index represents the reading word number of all the historical news read by the user i in unit time, and describes the average reading speed of the user i. As shown in fig. 2, the reading speed of the user can reflect the interest degree of the user in the current news, and the personalized reading speedThe comparison between the average reading speed of the user i reading j news and the average reading speed of the user i can be measured, and when the ratio is larger than 1, the fact that the reading speed of the user is higher than the average reading speed and possibly hurried, and the user is interested in the reading speed is shownGenerally, when the ratio is less than 1, it indicates that the reading speed of the user is slow, indicating that the user has a certain interest in it. Will personalize reading speedAs one of the weights of the user's interest level in news, the embedded representation of the current user can be obtained:
for a given user i, an embedded representation e (i) and an embedded representation e (t) of a candidate news jj) Through similarity calculation, the matching probability of the candidate news j and the user i can be calculated as follows: d (e), (i), e (t)j) And because news has strong timeliness, the release time of the news also should be used as an important standard for recommending, a news timeliness coefficient alpha is selected, alpha is more than or equal to 0 and less than or equal to 1, and then the probability of predicting that the user i clicks the news j is as follows:
since 85% of news articles will not be displayed on the news homepage after two days since the first appearance, the time efficiency coefficient α for news has a value range of:
αtime of release<2 days=1;α2 days<Time of release<5 days=0.15;αTime of release>5 days=0 (7)
In the step (5), for news with a reading time of 0, that is, recommended to the user without clicking, a spy algorithm is adopted, the news clicked and browsed by the user i is used as a positive sample set P, a part of subset S is randomly selected from the positive sample set P, S is added into the un-clicked news set to be used as a negative sample set N, then an SVN classification model is trained, the samples in the negative sample set N are scored by a trained classifier, and an average score of the spy set S is calculated after scoring.
According to the method, based on the time information characteristics of the user when reading the news, the recommendation weight of the news with too short reading time is reduced, and some news clicked by the user by mistake are filtered, so that the user preference model is more accurate; secondly, the value of the news timeliness coefficient is determined according to the release time of the news, the outdated news recommendation weight is reduced, and finally, for the news which is not clicked by the user, the interest degree of the user is judged by adopting a spy algorithm, so that the recommendation system is prevented from being divided into negative samples, a phenomenon that only part of hot spots are recommended in subsequent recommendations occurs, and the diversity of the recommendation system can be improved in the process.
The above embodiment is only one preferred embodiment of the present invention, and the protection scope of the present invention is not limited thereby, and any insubstantial changes and substitutions made by those skilled in the art based on the present invention are within the protection scope of the present invention.
Claims (6)
1. A news recommendation method based on user reading time behaviors is characterized in that: the method comprises the following steps:
extracting various information of news read by a user from a behavior log of the user, wherein the information comprises news titles, reading duration and news release time;
step (2) for each piece of extracted news, adopting a convolutional neural network to construct a personalized representation mode of the news;
in the step (3), because the interest degrees of the users to each piece of browsed news are different, different news are endowed with different weights through an attention mechanism;
step (4) weighting and aggregating all news read by the user within a period of time to represent the interest direction of the user, and taking the weighted and aggregated news as the characteristic representation of the user;
step 5, finding out news which is potentially interesting to the user by adopting a spy algorithm for news which is exposed but not clicked by the user, and adding the news which is generated newly and not recommended into candidate news to be recommended together with the news to be recommended for processing;
and (6) calculating the similarity between the candidate news to be recommended and the user characteristics, and recommending the news with higher similarity.
2. The news recommendation method based on the user reading time behavior as claimed in claim 1, wherein: in the step (1), extracting various information of news read by a user from a behavior log of the user to obtain the click history of the user i; each log comprises information such as a time stamp, a user ID, a news title and the like, and the click history of the user is represented asWhereinIs the jth news headline clicked by user i, N is the total number of news clicked by user i, t ═ w1,w2,w3,......]Title, w, representing newsiFor a word in a news headline, the news information is preliminarily extracted and correspondingly preprocessed.
3. The news recommendation method based on the user reading time behavior as claimed in claim 1, wherein: in the step (2), the news headline t in the step (1) is converted into an embedded matrix W through a word vector model, and the sub-matrix W of the W is subjected toi,i+l-1Performing convolution operation to obtain characteristicsAfter performing convolution operation on each sub-matrix, a feature map can be obtained:
wherein h is a convolution kernel, l is the size of the convolution kernel, f is a nonlinear function, a plurality of features are obtained by using a plurality of convolution kernels with different sizes, and finally the features are connected in series to form a final news representation vector:
e(t)=[Ch1 Ch2 ...... Chm] (2)。
4. the news recommendation method based on the user reading time behavior as claimed in claim 1, wherein: in the step (3), because the user has different interest degrees in different news, the weight of the different news occupying the user preference needs to be calculated, and an attention mechanism is used for calculation; representing the historical news in the step (2) into a vectorAs input, with the current candidate news tjRepresents the vector e (t)j) Calculating normalized influence weight:
the index describes the influence degree of the historical news read by the user on the news to be recommended.
5. The news recommendation method based on the user reading time behavior as claimed in claim 1, wherein: in the step (4), in order to express the interest degree of the user in different news, introducing personalized reading speed:
wherein the content of the first and second substances,indicating the number of words read by the user i per unit time when reading the j news, the index describing the reading speed of the user i when reading the j news,the index represents the reading word number of all historical news read by the user i in unit time, and the index describes the average reading speed of the user i; the user reading speed can reflect the interest degree of the user in the current news and personalized reading speedComparing the average reading speed and the average reading speed when the user i reads the j news; will personalize reading speedAs one of the weights of the user's interest level in news, the embedded representation of the current user can be obtained:
for a given user i, an embedded representation e (i) and an embedded representation e (t) of a candidate news jj) And calculating the matching probability of the candidate news j and the user i through similarity calculation, wherein the matching probability is as follows: d (e), (i), e (t)j) And because news has strong timeliness, the release time of the news also should be used as an important standard for recommending, a news timeliness coefficient alpha is selected, alpha is more than or equal to 0 and less than or equal to 1, and then the probability of predicting that the user i clicks the news j is as follows:
since 85% of news articles will not be displayed on the news homepage after two days since the first appearance, the time efficiency coefficient α for news has a value range of:
αtime of release<2 days=1;α2 days<Time of release<5 days=0.15;αTime of release>5 days=0 (7)。
6. The news recommendation method based on the user reading time behavior as claimed in claim 1, wherein: in the step (5), for news with a reading time length of 0, namely, news recommended to a user without a click, a spy algorithm is adopted, the news clicked and browsed by the user i is used as a positive sample set P, a part of subset S is randomly selected from the positive sample set P, S is added into the un-clicked news set to be used as a negative sample set N, then an SVN classification model is trained, the samples in the negative sample set N are scored by a trained classifier, an average score of the spy set S is calculated after scoring, as the set S is the news clicked by the user, the samples with the score lower than the score in the negative sample set N are divided into negative samples, and the samples with the score higher than the score are not divided into negative samples; .
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110258628.1A CN113065062A (en) | 2021-03-09 | 2021-03-09 | News recommendation method and system based on user reading time behavior |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110258628.1A CN113065062A (en) | 2021-03-09 | 2021-03-09 | News recommendation method and system based on user reading time behavior |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113065062A true CN113065062A (en) | 2021-07-02 |
Family
ID=76560443
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110258628.1A Pending CN113065062A (en) | 2021-03-09 | 2021-03-09 | News recommendation method and system based on user reading time behavior |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113065062A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113434779A (en) * | 2021-07-22 | 2021-09-24 | 咪咕数字传媒有限公司 | Interactive reading method and device capable of intelligent recommendation, computing equipment and storage medium |
CN113590806A (en) * | 2021-08-02 | 2021-11-02 | 山东建筑大学 | Personalized news recommendation method and system based on object three-dimensional language concept |
CN115048586A (en) * | 2022-08-11 | 2022-09-13 | 广东工业大学 | Multi-feature-fused news recommendation method and system |
CN115098767A (en) * | 2022-05-29 | 2022-09-23 | 北京理工大学 | News recommendation method based on interest perception and user similarity |
CN116881574A (en) * | 2023-09-07 | 2023-10-13 | 中科数创(北京)数字传媒有限公司 | Directional science popularization pushing method and system based on user portrait |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104166668A (en) * | 2014-06-09 | 2014-11-26 | 南京邮电大学 | News recommendation system and method based on FOLFM model |
US20170132230A1 (en) * | 2015-11-09 | 2017-05-11 | WP Company LLC d/b/a The Washington Post | Systems and methods for recommending temporally relevant news content using implicit feedback data |
CN106682152A (en) * | 2016-12-23 | 2017-05-17 | 西安电子科技大学 | Recommendation method for personalized information |
CN106777139A (en) * | 2016-12-19 | 2017-05-31 | 浙江工业大学 | User based on reading time reads the personalized push method of preference statistics |
CN109492157A (en) * | 2018-10-24 | 2019-03-19 | 华侨大学 | Based on RNN, the news recommended method of attention mechanism and theme characterizing method |
CN110188208A (en) * | 2019-06-04 | 2019-08-30 | 河海大学 | A kind of the information resources inquiry recommended method and system of knowledge based map |
CN111639258A (en) * | 2020-05-20 | 2020-09-08 | 苏州遐迩信息技术有限公司 | News recommendation method based on neural network |
-
2021
- 2021-03-09 CN CN202110258628.1A patent/CN113065062A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104166668A (en) * | 2014-06-09 | 2014-11-26 | 南京邮电大学 | News recommendation system and method based on FOLFM model |
US20170132230A1 (en) * | 2015-11-09 | 2017-05-11 | WP Company LLC d/b/a The Washington Post | Systems and methods for recommending temporally relevant news content using implicit feedback data |
CN106777139A (en) * | 2016-12-19 | 2017-05-31 | 浙江工业大学 | User based on reading time reads the personalized push method of preference statistics |
CN106682152A (en) * | 2016-12-23 | 2017-05-17 | 西安电子科技大学 | Recommendation method for personalized information |
CN109492157A (en) * | 2018-10-24 | 2019-03-19 | 华侨大学 | Based on RNN, the news recommended method of attention mechanism and theme characterizing method |
CN110188208A (en) * | 2019-06-04 | 2019-08-30 | 河海大学 | A kind of the information resources inquiry recommended method and system of knowledge based map |
CN111639258A (en) * | 2020-05-20 | 2020-09-08 | 苏州遐迩信息技术有限公司 | News recommendation method based on neural network |
Non-Patent Citations (2)
Title |
---|
吴彦文 等: "一种基于改进型协同过滤算法的新闻推荐系统", 计算机工程与科学, no. 06 * |
温宇俊;袁晖;: "基于用户上下文序列的个性化新闻推荐方法研究", 中国传媒大学学报(自然科学版), no. 04 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113434779A (en) * | 2021-07-22 | 2021-09-24 | 咪咕数字传媒有限公司 | Interactive reading method and device capable of intelligent recommendation, computing equipment and storage medium |
CN113590806A (en) * | 2021-08-02 | 2021-11-02 | 山东建筑大学 | Personalized news recommendation method and system based on object three-dimensional language concept |
CN113590806B (en) * | 2021-08-02 | 2022-05-27 | 山东建筑大学 | Personalized news recommendation method and system based on object three-dimensional language concept |
CN115098767A (en) * | 2022-05-29 | 2022-09-23 | 北京理工大学 | News recommendation method based on interest perception and user similarity |
CN115048586A (en) * | 2022-08-11 | 2022-09-13 | 广东工业大学 | Multi-feature-fused news recommendation method and system |
CN116881574A (en) * | 2023-09-07 | 2023-10-13 | 中科数创(北京)数字传媒有限公司 | Directional science popularization pushing method and system based on user portrait |
CN116881574B (en) * | 2023-09-07 | 2023-11-28 | 中科数创(北京)数字传媒有限公司 | Directional science popularization pushing method and system based on user portrait |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108629665B (en) | Personalized commodity recommendation method and system | |
US10366119B2 (en) | Customized content stream utilizing dwelltime-based machine learning | |
CN113065062A (en) | News recommendation method and system based on user reading time behavior | |
US11698939B2 (en) | Prioritizing items from different categories in a news stream | |
CN105224699B (en) | News recommendation method and device | |
CN108550068B (en) | Personalized commodity recommendation method and system based on user behavior analysis | |
CN111709812A (en) | E-commerce platform commodity recommendation method and system based on user dynamic classification | |
CN109064285B (en) | Commodity recommendation sequence and commodity recommendation method | |
CN109918563B (en) | Book recommendation method based on public data | |
CN105574216A (en) | Personalized recommendation method and system based on probability model and user behavior analysis | |
CN105426528A (en) | Retrieving and ordering method and system for commodity data | |
US8825641B2 (en) | Measuring duplication in search results | |
CN111724235A (en) | Online commodity recommendation method based on user novelty | |
CN116431931B (en) | Real-time incremental data statistical analysis method | |
CN111914172A (en) | Medical information recommendation method and system based on user tags | |
CN111651678B (en) | Personalized recommendation method based on knowledge graph | |
CN116541607B (en) | Intelligent recommendation method based on commodity retrieval data analysis | |
Han | Personalized news recommendation and simulation based on improved collaborative filtering algorithm | |
CN112288554B (en) | Commodity recommendation method and device, storage medium and electronic device | |
CN107545457B (en) | Automobile racing product type determination method and device | |
KR20030058660A (en) | The method of Collaborative Filtering using content references of users in Personalization System | |
CN113592588B (en) | E-commerce platform commodity recommendation system and method based on big data collaborative filtering technology | |
US20150142782A1 (en) | Method for associating metadata with images | |
CN113326432A (en) | Model optimization method based on decision tree and recommendation method | |
US12001497B2 (en) | Prioritizing items from different categories in a news stream |
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 |