CN105869058B - A kind of method that multilayer latent variable model user portrait extracts - Google Patents
A kind of method that multilayer latent variable model user portrait extracts Download PDFInfo
- Publication number
- CN105869058B CN105869058B CN201610250016.7A CN201610250016A CN105869058B CN 105869058 B CN105869058 B CN 105869058B CN 201610250016 A CN201610250016 A CN 201610250016A CN 105869058 B CN105869058 B CN 105869058B
- Authority
- CN
- China
- Prior art keywords
- user
- collection
- model
- description information
- interest
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 10
- 239000000284 extract Substances 0.000 title abstract 2
- 238000009826 distribution Methods 0.000 claims abstract description 37
- 238000012360 testing method Methods 0.000 claims description 8
- 238000010586 diagram Methods 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 5
- 150000001875 compounds Chemical class 0.000 claims description 3
- 230000009193 crawling Effects 0.000 claims 1
- 238000000605 extraction Methods 0.000 abstract description 4
- 238000007418 data mining Methods 0.000 abstract description 2
- 230000001174 ascending effect Effects 0.000 abstract 1
- 238000004364 calculation method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 239000006185 dispersion Substances 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012797 qualification Methods 0.000 description 2
- 235000001674 Agaricus brunnescens Nutrition 0.000 description 1
- 238000003482 Pinner synthesis reaction Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000003796 beauty Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000003012 network analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
-
- 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
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
A kind of method that multilayer latent variable model user portrait extracts, is related to data mining and recommender system field.The present invention is towards social plan exhibition network extraction user's portrait, and the data of the both modalities which of user behavior, propose a kind of user's portrait extracting method of multilayer latent variable model in the text description information and forwarding chain for collection entry.It introduces LDA model and the implicit theme distribution of user is obtained to text description information, being distributed to theme interest for theme distribution is implied based on user;Theme distribution is implied in conjunction with user and theme interest is distributed to obtain the interest distribution of user.User's socialization community is found based on multilayer latent variable model, and obtains user's recommendation results in conjunction with Jensen-Shannon divergence ascending sort.The present invention carries out user's socialization community discovery using the information of user version information and forwarding two kinds of different modalities of chain user behavior, realizes that user recommends.
Description
Technical Field
The invention relates to the field of data mining and recommending systems, in particular to research and implementation of a multi-layer latent variable model user portrait extraction method.
Background
Social media refers to a series of web applications that are built on the technical and conscious aspects of web2.0 and allow users to create and communicate content themselves. Since 2009, some professional social curation networks (e.g., Pinterest, snip. it, zoopit, petal nets, etc.) have formally emerged. The term "social curation" refers to the pronoun of people's activities such as collecting, organizing, and sharing information over the network. Traditional social networks are user-centric, and social curation networks are content-centric. The social strategy network is guided by user interests, users can create contents by themselves and collect contents concerned by themselves in other websites into their directories by linking and copying, and the users can sort and sort the personal collections according to their interests, so that the social strategy network provides very convenient network linking, network resource collection, sorting and sharing functions for the users. Other users can comment, like and change the collection of the user. The convenient publishing function enables the user to truly realize one-click expression and conveniently share own viewpoints. Pinterest is one of the websites that first integrated one-click content authoring and structuring into a content management set (called "Board"). The rapid development of Pinterest fully accounts for the appeal of such social curated networks to mass users. From 2012 onwards, more than ten Pinterest-like social curation networks such as petal nets, beauty, mushroom streets, etc. have appeared in China.
Extracting personalized user portraits on social networks is a key technology. There are many ways to extract user portraits for social networks, but most focus on predicting user attribute information, such as demographic characteristics, gender, age, religion, etc. Predictive models are often built on a series of feature sets (often natural language textual content) built by users, rather than a systematic model for recommendations or community discovery. User portrait extraction based on latent variable models is applied to text content analysis, traditional recommendation systems, social recommendation systems and social network analysis.
Social curation networks contain different kinds of information: the collection items, the collection book and the collection classification, short text information of the collection items generated by the user, information of a forwarding chain, attention information and the like. The user is composed of a series of collection item sets arranged in a collection book, and one collection item contains rich information, such as description information, information of the relevant collection book, original information forwarded by the collection item and the like. Social curation of information across multiple modalities of a network presents challenges to user modeling. How to carry out latent variable modeling on the description text information of the collection items and the user forwarding chain behaviors, find a potential user community and recommend users similar to the target user interests are the key points of the invention.
Disclosure of Invention
The user portrait extraction method based on the multi-modal latent variable modeling is researched for a social strategy development network and aiming at the characteristic that user data modes are various, and is used for community discovery and user recommendation realization.
In order to realize the problem, the invention provides a double-layer Bayesian latent variable model for describing the user. The method comprises the following steps:
A. and establishing a word bank and a stop word bank, and performing word segmentation on the text description information of all the collection items of the user by using a word segmentation tool ICTCCLAS.
B. In the social policy network, the favorite items of the current user can be collected by other users. If a user changes to a collection item, all self information except comments, praise and the like of the collection item is copied into a self collection book.
Collecting all collection item forwarding chain data of a target user set, and acquiring data from each collection item to an original collection item according to data of a person who forwards the collection item in the collection items. Data is crawled to the parent level starting with the current collection entry. Taking as a guide where this data is forwarded from, we trace back to the original favorites entry location. Each node in the tracing process is a copy of the original collection entry, and a chain-shaped path diagram, called a forwarding chain, is formed by the nodes. Each forwarding chain is made up of a collection of collection entries. One node on the forwarding chain is represented by the creating user ID for each forwarding favorites entry.
C. The multi-layer Latent variable model is based on a popular Bayesian Latent variable model, namely, Late Dirichlet Allocation (LDA), and a user interest model is extracted from a set (represented by user ID, and referred to as "pinners" herein) formed by all text information and forwarding chain data of a target user set.
Further, the step C specifically includes:
c1, the first layer model is the implicit theme of the text description information of the user collection item.
C2, calculating the probability of each collection item belonging to each topic
C3, the second layer model is to calculate the interest distribution of the implied subject.
C4, calculating the interest distribution of the user.
Further, the step C1 includes:
c11 and LDA are popular Bayesian latent variable models and are widely applied to machine learning and natural language processing. The basic idea of LDA is that a document is a collection of multiple topics, one topic being the probability distribution of words. LDA is based on the "bag of words" assumption, that is, the order of words in a document can be ignored. The user text information is subjected to LDA model, and the user and the characters are associated through the implied subject to generate a user-subject-word three-layer Bayesian model.
C12 and Perplexity are used as a metric LDA model, and the smaller the Perplexity value is, the better the model is. Measuring the first-layer LDA model by Perplexity, calculating the Perplexity model, and selecting the best number of subjects NT1Used in the next step.
Wherein, UtestTo test a set of users, UtTotal number of users for test set, wuSet of words, p (w), describing information for user u's collection entryu) Generating a probability, N, for a set of words of the collection entry description information of user u under a user modeluThe collection entry for user u describes the total number of words of the set of words of information. K is the total number of topics, NmA set of words of item description information is collected for all users.
The step C2 includes:
c21 description information of a favorite is created by user or transferred, composed of a group of words, and can be represented as wpin={w1,w2,...wi...,wNIn which wiFor the ith word of the collection item description information, N is the collection item description informationAnd the total number of the word set. Topic collectionsIs the probability distribution of a word, where zkFor the kth topic, NT1Is the best overall theme. By usingIndicating that a Collection entry pin belongs to a topic zkThe probability of (c).
C22, calculating each collection item pin belonging to a subject z according to the result of the first-layer LDA modelkProbability of (2)According to the pareto law, we choose 0.2 as the threshold ifThe collection entry belongs to this topic. Thus obtaining a compound of NT1A collection of collection entries for individual topics.
The step C3 includes:
c31, a favorites entry is typically from one user to the end of another. A favorites entry may also be forwarded by other users, with a forward chain being the user's propagation through the social curation network. Each collection entry is represented by a user ID on the forwarding chain.
C32, treating the user ID as an independent word, according to the theory that LDA is based on the "bag of words" assumption. Thus, for implied subject NT1And obtaining a three-layer Bayesian model of a theme-interest-user ID through the LDA model, and obtaining interest distribution of the hidden description information theme.
The step C4 includes:
and C41, combining the user implicit theme distribution and the interest distribution of the theme, and obtaining the interest distribution of the user through matrix multiplication. It can be said that a user is a collection of interests, one interest being the probability distribution of the user ID.
D. User recommendation based on multi-layer latent variable model
The step D comprises the following steps:
d1, user u1And user u2Similarity and user u2And user u1The similarity between them is the same. Jensen-Shannon divergence is a method of measuring the distance (degree of similarity) of probability distributions. Jensen-Shannon divergence is more balanced than Kullback-Leibler divergence, and the order of the parameters is symmetrical, i.e., the results are independent of the order of the parameters. The smaller the Jensen-Shannon divergence, the greater the similarity. And calculating the Jensen-Shannon divergence of the target user and all users on the forwarding chains as the similarity between the two users P and Q.
Wherein,
the user set on all collection item forwarding chains is called other user set Urec={uR1,uR2,…,uRi,…,uRNIn which uRiThe node number of the collection entry set is the Ri-th user ID of the node on the forwarding chain, and the RN is the node number of the collection entry set on the forwarding chain. For the target user set, calculating each target user uiAnd RN other user sets UrecThe Jensen-Shannon divergence of each user is used as the similarity between two users, and the calculation result can be expressed asWhereinRepresenting a target user uiFor user uRiJensen-Shannon dispersion value.
D2 for eyesTarget user uiCalculated Jensen-Shannon divergence valueThe smaller the value, the more similar the user's interest, sorted from small to large. And taking Top-N as a recommended user to the target user.
Description of the drawings:
fig. 1 is a schematic diagram of a forwarding chain according to the present example.
FIG. 2 is a schematic diagram of a multi-layer latent variable model framework according to the present example.
FIG. 3 is a diagram of the results of a Perplexity of the present example.
FIG. 4 is a diagram illustrating the results of a community discovery process according to this example.
FIG. 5 is a diagram illustrating a MAP result according to the example.
The specific implementation mode is as follows:
the technical solution of the present invention will be described in more detail with reference to the accompanying drawings and examples.
The embodiment is performed for real data of a certain social curation network, in the example, 100 target users are real users in the network and respectively come from three classifications, where nos. 1 to 35 belong to classification one, nos. 36 to 75 belong to classification two, nos. 76 to 100 belong to classification three, and the total includes 633337 collection entries and forwarding chains corresponding to the collection entries.
A. A new word bank is established, which contains about 300000 new words, and is a common and popular keyword. A stop word thesaurus is established containing 1433 stop words, which have no specific meaning in the sentence.
The description information of one collection entry in the example is 'Taobao commodity description must have which qualifications?', and is subjected to word segmentation to obtain { Taobao, commodity, description, possess, qualifications }, and { must, which } stop words, so that the words are removed.
B. Read-in target user collection entry forwardingAnd chain data, wherein each piece of forwarding chain data is marked by a user ID (serial number) of a creating user of each node collection entry on the forwarding chain, and is represented as R ═ { p { (p) }1,p2,……,pn}. The last two nodes p of each forwarding chainn-1And pnAnd (5) removing. The forwarding chain for a favorites entry for a target user in this example may be represented by a user ID as {38450, 115078, 86804, 60952, 310115, 86588, 269584, 280741, 298423, 15278, 31028, 256217, 271691}, with node p removedn-1And pnThe postforward chain is denoted by 86804, 60952, 310115, 86588, 269584, 280741, 298423, 15278, 31028, 256217, 271691.
C. And extracting the user interest model based on the text information and the forwarding chain data.
The step C specifically comprises the following steps:
c1, calculating the hidden theme of the text description information of the user collection items by the first layer model, and selecting the optimal theme number.
C2, calculating the probability of each collection item belonging to each topic
C3, the second layer model is to calculate the interest distribution of the implied subject.
C4, calculating the interest distribution of the user.
The step C1 specifically includes:
and C11, performing LDA modeling on the description information of 100 target users to obtain a user-theme-word three-layer Bayesian model, and associating the users and words through themes.
Where p (w | u) represents the word distribution of the 100 target user sets, p (w | t) andtopic distribution, p (t | u) and θ, representing a set of wordsuRepresenting the probability distribution of the occurrence of a topic in the set of users.
C12, 10% of the data set selected in the experimentAs a test set, i.e. 10 users as a test set. Calculating Perplexity when the number of themes NT1Perplexity values tend to plateau at > 30, indicating NT1The model is optimal in quality and computational complexity of more than or equal to 30. So set up NT130. One user u out of the 100 target users in this exampleiFor 30 subjects the probability isWherein
The step C2 specifically includes:
c21, calculating the result of three-layer Bayes model of the first layer user-subject-word, each collection item pin belonging to a subject zkProbability of (2)The description information of a collection item pin is divided into words { Taobao, commodity, description, possession and quality }, wherein each word is corresponding to a subject zkThe probabilities of (1) are {0.805, 0.456, 0.771, 0.002, 0.002}, respectively, then
C22, calculating that all collection items of 100 target user sets belong to NT1Probability of 30 topicsIf it is notThe collection entry is considered to belong to this topic. Thus obtaining a compound of NT1A collection of 30 subject collection entries.
The step C3 includes:
c31, each favorites entry is represented by a user ID on the forwarding chain. N is a radical ofT1The 30 topics are represented as a set of related forwarding chains, and also as a set of related user IDs of nodes on the forwarding chains.
C32, to NT1And obtaining a three-layer Bayesian model of a subject-interest-user ID through the LDA model for 30 subjects to obtain the interest distribution of the implicitly described information subject, and associating the subject with the user ID through the interest.
Wherein p (uid | t) represents NT1User ID distribution of 30 topics, p (uid | int) anddistribution of interest, p (int | t) and θ, representing a set of user IDstRepresenting the probability distribution of the occurrence of interest in the topic collection.
The step C4 includes:
c41, combining the user implicit topic distribution and the topic interest distribution, the implicit topic of the 100 target user sets can be expressed as: a matrix of 100 users by 30 topics, the interest distribution of which may be expressed as: matrix of 30 subjects interest. Through matrix multiplication, a matrix of 100 users' interests, i.e., a user interest probability distribution, can be calculated. It can also be said that a user is a collection of multiple interests, one interest being the probability distribution of the user ID.
p(int|u)=p(int|t)p(t|u)=θtθu
Where p (int | d) represents the probability distribution of the occurrence of interest among 100 users.
One advantage of the multi-modal latent variable model is that a community of users can be discovered from the description information and forwarding chain. Compared with a classic LDA model algorithm, the multi-mode latent variable model can better describe the user social community compared with the traditional LDA model algorithm.
D. User recommendation based on multi-layer latent variable model
The step D comprises the following steps:
d1, calculating each target user u for 100 users in the target user setiAnd RN other user sets UrecThe Jensen-Shannon divergence of each user is used as the similarity between two users, and the calculation result can be expressed asWhereinRepresenting a target user uiFor user uRiJensen-Shannon dispersion value.
D2, for target user uiCalculated Jensen-Shannon divergence valueIn order from small to large, the smaller the Jensen-Shannon divergence value is, the more similar the user interest distribution is. And taking Top-N as a recommended user to the target user. In this example, the present algorithm is compared to five baseline algorithms: (1) a recommendation algorithm based on a multi-modal latent variable model and Kullback-Leibler divergence, (2) a recommendation algorithm based on an LDA model and Jensen-Shannon divergence, (3) a recommendation algorithm based on an LDA model and Kullback-Leibler divergence, and (4) a URSRP algorithm, and (5) a popularity recommendation algorithm of user ID occurrence frequency. Compared with the average accuracy (MAP) index, the recommendation effect is obviously improved, and the MLLDA-JSD is improved compared with other baseline algorithms. We also verified that Jensen-Shannon divergence measures similarity between different users better than Kullback-Labler divergence.
Claims (1)
1. A method for extracting a user portrait of a multi-layer latent variable model is characterized by comprising the following steps:
A. establishing a word bank and a stop word bank, and performing word segmentation on the text description information of all the collection items of the user by using a word segmentation tool ICTCCLAS;
B. in the social strategy network, the collection items of the current user can be collected by other users; if a user changes to collect one collection item, all self information of the collection item is copied into a self collection book;
collecting all collection item forwarding chain data of a target user set, and acquiring data from each collection item to an original collection item according to data of a person who forwards the collection item in the collection items; crawling data to the parent level from the current collection item; taking the data of which the data is forwarded from as a guide, and tracing back to the position of the original collection item; each node in the tracing process is a copy of the original collection entry, and a chain-shaped path diagram is formed by the nodes and is called as a forwarding chain; each forwarding chain is composed of a set containing a plurality of collection items; representing a node on the forwarding chain by the creating user ID of each forwarded collection entry;
C. extracting a user interest model from a set formed by all text information and forwarding chain data of a target user set;
the step C specifically comprises the following steps:
c1, calculating the implicit theme of the text description information of the user collection item;
c2, calculating the probability that each collection item belongs to each topic;
c3, calculating interest distribution of the implied subject by the second-layer model;
c4, calculating the interest distribution of the user;
D. user recommendations based on a multi-layer latent variable model;
the step C1 includes:
c11, the user text information is associated with the user and the characters through the implicit theme by an LDA model to generate a user-theme-word three-layer Bayes model;
c12, measuring the first-layer LDA model by Perplexity, calculating the Perplexity model, and selecting the best number of subjects NT1Used in the next step;
wherein, UtestTo test a set of users, UtTotal number of users for test set, wuSet of words, p (w), describing information for user u's collection entryu) Generating a probability, N, for a set of words of the collection entry description information of user u under a user modeluTotal number of words of the word set of the collection item description information for user u; k is the total number of topics, NmCollecting a word set of item description information for all users;
the step C2 includes:
c21 description information of a favorite is created by user or transferred, and is composed of a group of words, denoted as wpin={w1,w2,…wi…,wNIn which wiThe ith word of the collection item description information is represented, and N is the total number of collection item description information word sets; topic collectionsIs the probability distribution of a word, where zkFor the kth topic, NT1The number of subjects is the best; by usingIndicating that a Collection entry pin belongs to a topic zkThe probability of (d);
c22, calculating each collection item pin belonging to a subject z according to the result of the first-layer LDA modelkProbability of (2)Selecting 0.2 as the critical value ifThe collection entry belongs to this topic; thus obtaining a compound of NT1A collection of collection items for the best topic;
the step C3 includes:
c31, each collection item is represented by a user ID on the forwarding chain;
c32, obtaining a three-layer Bayesian model of a subject-interest-user ID through an LDA model for the implied subject to obtain the interest distribution of the implied description information subject;
the step C4 includes:
and C41, combining the user implicit theme distribution and the interest distribution of the theme, and obtaining the interest distribution of the user through matrix multiplication.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610250016.7A CN105869058B (en) | 2016-04-21 | 2016-04-21 | A kind of method that multilayer latent variable model user portrait extracts |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610250016.7A CN105869058B (en) | 2016-04-21 | 2016-04-21 | A kind of method that multilayer latent variable model user portrait extracts |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105869058A CN105869058A (en) | 2016-08-17 |
CN105869058B true CN105869058B (en) | 2019-10-29 |
Family
ID=56632428
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610250016.7A Expired - Fee Related CN105869058B (en) | 2016-04-21 | 2016-04-21 | A kind of method that multilayer latent variable model user portrait extracts |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105869058B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107357835B (en) * | 2017-06-22 | 2020-11-03 | 电子科技大学 | Interest prediction mining method and system based on topic model and forgetting rule |
CN108876643A (en) * | 2018-05-24 | 2018-11-23 | 北京工业大学 | It is a kind of social activity plan exhibition network on acquire(Pin)Multimodal presentation method |
CN110209875B (en) * | 2018-07-03 | 2022-09-06 | 腾讯科技(深圳)有限公司 | User content portrait determination method, access object recommendation method and related device |
CN112836507B (en) * | 2021-01-13 | 2022-12-09 | 哈尔滨工程大学 | Method for extracting domain text theme |
CN116383521B (en) * | 2023-05-19 | 2023-08-29 | 苏州浪潮智能科技有限公司 | Subject word mining method and device, computer equipment and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103064917A (en) * | 2012-12-20 | 2013-04-24 | 中国科学院深圳先进技术研究院 | Specific-tendency high-influence user group discovering method orienting microblog |
CN103500340A (en) * | 2013-09-13 | 2014-01-08 | 南京邮电大学 | Human body behavior identification method based on thematic knowledge transfer |
CN103886067A (en) * | 2014-03-20 | 2014-06-25 | 浙江大学 | Method for recommending books through label implied topic |
CN104991956A (en) * | 2015-07-21 | 2015-10-21 | 中国人民解放军信息工程大学 | Microblog transmission group division and account activeness evaluation method based on theme possibility model |
CN105069003A (en) * | 2015-06-15 | 2015-11-18 | 北京工业大学 | User focus object recommendation calculation method based on forward chain similarity |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6981040B1 (en) * | 1999-12-28 | 2005-12-27 | Utopy, Inc. | Automatic, personalized online information and product services |
-
2016
- 2016-04-21 CN CN201610250016.7A patent/CN105869058B/en not_active Expired - Fee Related
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103064917A (en) * | 2012-12-20 | 2013-04-24 | 中国科学院深圳先进技术研究院 | Specific-tendency high-influence user group discovering method orienting microblog |
CN103500340A (en) * | 2013-09-13 | 2014-01-08 | 南京邮电大学 | Human body behavior identification method based on thematic knowledge transfer |
CN103886067A (en) * | 2014-03-20 | 2014-06-25 | 浙江大学 | Method for recommending books through label implied topic |
CN105069003A (en) * | 2015-06-15 | 2015-11-18 | 北京工业大学 | User focus object recommendation calculation method based on forward chain similarity |
CN104991956A (en) * | 2015-07-21 | 2015-10-21 | 中国人民解放军信息工程大学 | Microblog transmission group division and account activeness evaluation method based on theme possibility model |
Also Published As
Publication number | Publication date |
---|---|
CN105869058A (en) | 2016-08-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Giannoulakis et al. | Evaluating the descriptive power of Instagram hashtags | |
Lu et al. | GCAN: Graph-aware co-attention networks for explainable fake news detection on social media | |
Kumar et al. | Sentiment analysis of multimodal twitter data | |
CN105869058B (en) | A kind of method that multilayer latent variable model user portrait extracts | |
Sawant et al. | Automatic image semantic interpretation using social action and tagging data | |
Sun et al. | Leveraging content and connections for scientific article recommendation in social computing contexts | |
Pan et al. | Social media-based user embedding: A literature review | |
Cao et al. | BASS: A bootstrapping approach for aligning heterogenous social networks | |
Koggalahewa et al. | An unsupervised method for social network spammer detection based on user information interests | |
Faralli et al. | Automatic acquisition of a taxonomy of microblogs users’ interests | |
Shevade et al. | Modeling personal and social network context for event annotation in images | |
Kumar | Social Media Analytics for Stance Mining A Multi-Modal Approach with Weak Supervision. | |
Han et al. | Photos don't have me, but how do you know me? Analyzing and predicting users on Instagram | |
Peng et al. | HARSAM: A hybrid model for recommendation supported by self-attention mechanism | |
Lydiri et al. | A performant deep learning model for sentiment analysis of climate change | |
Sheeba et al. | A fuzzy logic based on sentiment classification | |
Chalehchaleh et al. | BRaG: a hybrid multi-feature framework for fake news detection on social media | |
Potha et al. | Dynamic ensemble selection for author verification | |
Zhao et al. | User-sentiment topic model: refining user's topics with sentiment information | |
Xia et al. | Content-irrelevant tag cleansing via bi-layer clustering and peer cooperation | |
Su et al. | Labeling faces with names based on the name semantic network | |
Liu et al. | Macro-scale mobile app market analysis using customized hierarchical categorization | |
Hannah et al. | A classification-based summarisation model for summarising text documents | |
Abulaish et al. | A layered approach for summarization and context learning from microblogging data | |
Sharma et al. | Determining the popularity of political parties using twitter sentiment analysis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20191029 |
|
CF01 | Termination of patent right due to non-payment of annual fee |