CN110110230A - A kind of recommended method to be scored based on user with comment - Google Patents
A kind of recommended method to be scored based on user with comment Download PDFInfo
- Publication number
- CN110110230A CN110110230A CN201910341581.8A CN201910341581A CN110110230A CN 110110230 A CN110110230 A CN 110110230A CN 201910341581 A CN201910341581 A CN 201910341581A CN 110110230 A CN110110230 A CN 110110230A
- Authority
- CN
- China
- Prior art keywords
- article
- comment
- user
- scoring
- data
- 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
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
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/205—Parsing
- G06F40/216—Parsing using statistical methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0282—Rating or review of business operators or products
Abstract
The invention discloses a kind of recommended methods to be scored based on user with comment, this method is to be pre-processed first by big data preconditioning technique to comment data, then article characteristics are extracted from pretreated user comment data using Word2Vec model, it is re-introduced into word frequency, scoring is commented on the time, and comment help degree improves article characteristics, the similarity between article finally is calculated using article characteristics, and carries out score in predicting and recommendation list generation.For the method for the present invention in user comment shortage of data, the description information by extracting article from the metadata of article, to fill the missing comment data of article, can preferably solve the problems, such as that article is cold-started as a user comment data.The predictablity rate of the method for the present invention is greatly improved, and can be realized and more accurately recommends.
Description
Technical field
The present invention relates to the technical fields of recommended method, refer in particular to a kind of recommendation side scored based on user with comment
Method.
Background technique
The high speed development of Internet technology significantly reduces the threshold that people obtain information, at the same time, global interconnection
Huge data all are being generated all the time on the net, human society has come into the epoch of information overload.Based on collaborative filtering
Personalized recommendation technology as solve problem of information overload effective means, it can by analyze user historical behavior,
The information such as goods attribute and context are excavated the information being consistent with user interest from the data of magnanimity and are recommended, not only
It helps people to improve the efficiency for obtaining valuable information, and information can be accurately presented in its interested user
In front, huge economic benefit is brought for enterprise.
In actual application, traditional Collaborative Filtering Recommendation Algorithm faces Deta sparseness and cold start-up problem, causes
The accuracy and operational efficiency of algorithm prediction are relatively low.For the deficiency for making up Collaborative Filtering Recommendation Algorithm, this paper presents one kind
The recommended method for being scored and being commented on based on user.
Summary of the invention
It is an object of the invention to overcome the deficiencies in the prior art, propose a kind of recommendation scored based on user with comment
Method can be realized and more accurately recommend, and this method passes through big data preconditioning technique first and pre-processes to comment data,
Then article characteristics are extracted from pretreated user comment data using Word2Vec model, is re-introduced into word frequency, scored, comment
It by the time, comments on help degree and improves article characteristics, finally calculate the similarity between article using article characteristics, and score
Prediction and recommendation list generate.
To achieve the above object, a kind of technical solution provided by the present invention are as follows: recommendation scored based on user with comment
Method, comprising the following steps:
Step 1, comment data pretreatment: including goods review polymerization, missing comment filling, data de-duplication and data
Format conversion；
Step 2, article characteristics are extracted: article spy is extracted from pretreated comment data using Word2Vec model
Sign；
Step 3 improves article characteristics: utilizing word frequency, scoring, comment time and the comment help degree for including in user comment
Information improves article characteristics；
Step 4 calculates article similarity: being calculated between article based on improved article characteristics using cosine similarity
Similarity；
Step 5, prediction article scoring: according to scoring of the similarity calculation user to article between article, user is obtained
Prediction scoring to article；
Step 6 generates recommendation list: generating initial recommendation list to the prediction scoring of article according to user, then mistake
Filter user has scored article, generates final recommendation list.
In step 1, the comment data pretreatment is using big data preconditioning technique to original user comment number
According to progress data cleansing conversion, comprising:
All user comments for describing same article: polymerizeing by goods review polymerization according to the unique identification of article,
Obtain the comment set of same article；
Missing comment filling: for there is no the article of user comment, retouching for article is extracted from the metadata of article
Information is stated as a user comment data, to fill the missing comment data of article；
Data de-duplication: to all user comments in same goods review set, them are calculated using editing distance
Between similarity, the user comment high for similarity, be considered as repetition comment deleted；
Data Format Transform: to every user comment data, carrying out punctuation mark filtering, and the conversion of word capital and small letter segments,
Stem extracts and removal stop words processing.
In step 2, article characteristics extraction refers to using Word2Vec model from pretreated goods review number
According to middle extraction article characteristics, detailed process is as follows:
Step 2.1, term vector conversion: using Word2Vec model by the word of all user comments in same article set
Remittance is converted to the term vector of low-dimensional；
Step 2.2, comment feature vector conversion: the corresponding term vector of words all in corresponding comment is added up, then
Divided by the vocabulary quantity for including in this comment, the feature vector commented on；
Step 2.3, the conversion of article characteristics vector: it after the feature vector commented on, comments all in goods review set
It adds up by corresponding comment feature vector, then obtains the spy of article divided by the number of reviews for including in goods review set
Levy vector.
In step 3, described that article characteristics are improved with the word frequency referred to using including in user comment, scoring is commented
Article characteristics are improved by time and comment help degree information, detailed process is as follows:
Step 3.1 is introduced into word frequency improvement article characteristics: the word frequency of word in user comment is calculated using TF-IDF algorithm,
Article characteristics are improved according to word frequency weight；
Step 3.2 introduces scoring and improves article characteristics: assign higher weights to the higher user comment of scoring, scoring compared with
Low user comment assigns lower weight, is improved according to scoring weight to article characteristics；
Step 3.3 introduces comment time improvement article characteristics: the closer user comment of current time of adjusting the distance assigns higher
Weight is commented on apart from current time user farther out and assigns lower weight, is changed according to comment time weighting to article characteristics
Into；
Step 3.4 introduces comment help degree improvement article characteristics: assigning to the higher user comment of comment help degree higher
Weight is commented on apart from current time user farther out and assigns lower weight, is changed according to comment time weighting to article characteristics
Into；
In step 4, it is described calculate article similarity and refer to compare article two-by-two using cosine similarity measure
Feature vector obtains the similarity between article；
In steps of 5, the prediction article scoring, which refers to, comments article according to the similarity calculation user between article
Point, it obtains user and scores the prediction of article, detailed process is as follows:
Step 5.1 is ranked up from high to low according to the similarity between article, obtain with target item it is most like before
K article；
Step 5.2, the scoring according to user to this preceding k article predict scoring of the user to target item.
In step 6, the recommendation list, which generates, refers to the recommendation initial according to prediction scoring generation of the user to article
Then list filters user and has scored article, generates final recommendation list, detailed process is as follows:
Step 6.1, initial recommendation list generate: being ranked up, obtained from high to low to the prediction scoring of article according to user
To initial recommendation list；
Step 6.2, article filtering and sequence: user had given the article of scoring in filtering initial recommendation list, again
It is ranked up；
Step 6.3, consequently recommended list generate: it is raw to choose top n article for the article sorted lists in read step 6.2
At recommendation list.
Compared with prior art, the present invention have the following advantages that with the utility model has the advantages that
1, user comment content contains richer information than scoring, and comment content is usually text data, compared to letter
Single score value, can be as the foundation for explaining user's scoring, concern of the reflection user to feature in terms of article.
2, user comment content can be truer, embodiment article characteristics more precisely, the object given compared to scoring and businessman
Product metamessage can more relevantly reflect the article characteristics of user's concern.
3, user comment data can alleviate score data sparsity problem, when score data is sparse, can push away to collaboration
It recommends algorithm and causes serious influence, but as long as there is a small number of user comment contents, it will be able to establish article characteristics with this.
4, the method for the present invention in user comment shortage of data, believe by the description by extracting article from the metadata of article
Breath is used as a user comment data, to fill the missing comment data of article, can preferably solve the problems, such as that article is cold-started.
5, the predictablity rate of the method for the present invention is greatly improved, and can be realized and more accurately recommends.
Detailed description of the invention
Fig. 1 is the overall flow schematic diagram of the method for the present invention.
Fig. 2 is the pretreated specific flow chart of comment data in the method for the present invention.
Fig. 3 is the specific flow chart that article characteristics are extracted in the method for the present invention.
Specific embodiment
The present invention is further explained in the light of specific embodiments.
As shown in Figure 1 to Figure 3, the recommended method for scoring and commenting on based on user provided by the present embodiment, including it is following
Step:
Step 1, comment data pretreatment carry out data to original user comment data using big data preconditioning technique
Cleaning conversion, comprising:
All user comments for describing same article: polymerizeing by goods review polymerization according to the unique identification of article,
Obtain the comment set of same article；
Missing comment filling: for there is no the article of user comment, retouching for article is extracted from the metadata of article
Information is stated as a user comment data, to fill the missing comment data of article；
Data de-duplication: to all user comments in same goods review set, them are calculated using editing distance
Between similarity, the user comment high for similarity, be considered as repetition comment deleted；
Data Format Transform: to every user comment data, carrying out punctuation mark filtering, and the conversion of word capital and small letter segments,
Stem extracts and removal stop words processing.
Step 2, article characteristics are extracted: article spy is extracted from pretreated comment data using Word2Vec model
Sign, detailed process is as follows:
Step 2.1, term vector conversion: using Word2Vec model by the word of all user comments in same article set
Remittance is converted to the term vector of low-dimensional；
Step 2.2, comment feature vector conversion: the corresponding term vector of words all in corresponding comment is added up, then
Divided by the vocabulary quantity for including in this comment, the feature vector commented on；
Step 2.3, the conversion of article characteristics vector: it after the feature vector commented on, comments all in goods review set
It adds up by corresponding comment feature vector, then obtains the spy of article divided by the number of reviews for including in goods review set
Levy vector.
Step 3 improves article characteristics: utilizing word frequency, scoring, comment time and the comment help degree for including in user comment
Information improves article characteristics, and detailed process is as follows:
Step 3.1 is introduced into word frequency improvement article characteristics: the word frequency of word in user comment is calculated using TF-IDF algorithm,
Article characteristics are improved according to word frequency weight；
Step 3.2 introduces scoring and improves article characteristics: assign higher weights to the higher user comment of scoring, scoring compared with
Low user comment assigns lower weight, is improved according to scoring weight to article characteristics；
Step 3.3 introduces comment time improvement article characteristics: the closer user comment of current time of adjusting the distance assigns higher
Weight is commented on apart from current time user farther out and assigns lower weight, is changed according to comment time weighting to article characteristics
Into；
Step 3.4 introduces comment help degree improvement article characteristics: assigning to the higher user comment of comment help degree higher
Weight is commented on apart from current time user farther out and assigns lower weight, is changed according to comment time weighting to article characteristics
Into.
Step 4 calculates article similarity: being calculated between article based on improved article characteristics using cosine similarity
Similarity specifically compares the feature vector of article two-by-two using cosine similarity measure, obtains similar between article
Degree.
Step 5, prediction article scoring: according to scoring of the similarity calculation user to article between article, user is obtained
Prediction scoring to article, detailed process is as follows:
Step 5.1 is ranked up from high to low according to the similarity between article, obtain with target item it is most like before
K article；
Step 5.2, the scoring according to user to this preceding k article predict scoring of the user to target item.
Step 6 generates recommendation list: generating initial recommendation list to the prediction scoring of article according to user, then mistake
Filter user has scored article, generates final recommendation list, detailed process is as follows:
Step 6.1, initial recommendation list generate: being ranked up, obtained from high to low to the prediction scoring of article according to user
To initial recommendation list；
Step 6.2, article filtering and sequence: user had given the article of scoring in filtering initial recommendation list, again
It is ranked up；
Step 6.3, consequently recommended list generate: it is raw to choose top n article for the article sorted lists in read step 6.2
At recommendation list.
Embodiment described above is only the preferred embodiments of the invention, and but not intended to limit the scope of the present invention, therefore
All shapes according to the present invention change made by principle, should all be included within the scope of protection of the present invention.
Claims (7)
1. a kind of recommended method to be scored based on user with comment, which comprises the following steps:
Step 1, comment data pretreatment: including goods review polymerization, missing comment filling, data de-duplication and data format
Conversion；
Step 2, article characteristics are extracted: extracting article characteristics from pretreated comment data using Word2Vec model；
Step 3 improves article characteristics: utilizing word frequency, scoring, comment time and the comment help degree information for including in user comment
Article characteristics are improved；
Step 4 calculates article similarity: utilizing cosine similarity to calculate based on improved article characteristics similar between article
Degree；
Step 5, prediction article scoring: according to scoring of the similarity calculation user to article between article, user is obtained to object
The prediction of product is scored；
Step 6 generates recommendation list: generating initial recommendation list to the prediction scoring of article according to user, then filtering is used
Scored article at family, generates final recommendation list.
2. a kind of recommended method to be scored based on user with comment according to claim 1, it is characterised in that: in step 1
In, the comment data pretreatment, which refers to, carries out data cleansing to original user comment text using big data preconditioning technique
Conversion, comprising:
Goods review polymerization: all user comments for describing same article are polymerize according to the unique identification of article, are obtained
The comment set of same article；
Missing comment filling: for there is no the article of user comment, the description letter of article is extracted from the metadata of article
Breath is used as a user comment data, to fill the missing comment data of article；
Data de-duplication: it to all user comments in same goods review set, is calculated between them using editing distance
Similarity, the user comment high for similarity, be considered as repetition comment deleted；
Data Format Transform: to every user comment text, punctuation mark filtering, the conversion of word capital and small letter, participle, stem are carried out
It extracts and removal stop words is handled.
3. a kind of recommended method to be scored based on user with comment according to claim 1, it is characterised in that: in step 2
In, the article characteristics extraction refers to extracts article spy using Word2Vec model from pretreated goods review text
Sign, detailed process is as follows:
Step 2.1, term vector conversion: the vocabulary of all user comments in same article set is turned using Word2Vec model
It is changed to the term vector of low-dimensional；
Step 2.2, comment feature vector conversion: the corresponding term vector of all words will add up in corresponding comment, then divided by
The vocabulary quantity for including in this comment, the feature vector commented on；
Step 2.3, the conversion of article characteristics vector: after the feature vector commented on, to comments pair all in goods review set
The comment feature vector answered adds up, then obtained divided by the number of reviews for including in goods review set the feature of article to
Amount.
4. a kind of recommended method to be scored based on user with comment according to claim 1, it is characterised in that: in step 3
In, the improvement article characteristics, which refer to, utilizes the word frequency for including in user comment, scoring, comment time and comment help degree information
Article characteristics are improved, detailed process is as follows:
Step 3.1 is introduced into word frequency improvement article characteristics: the word frequency of word in user comment is calculated using TF-IDF algorithm, according to
Word frequency weight improves article characteristics；
Step 3.2 introduces scoring improvement article characteristics: to scoring, high user comment assigns high weight, and the low user that scores comments
By low weight is assigned, article characteristics are improved according to scoring weight；
Step 3.3 introduces comment time improvement article characteristics: the close user comment of current time of adjusting the distance assigns high weight, away from
The user comment remote from current time assigns low weight, is improved according to comment time weighting to article characteristics；
Step 3.4 introduces comment help degree improvement article characteristics: the user comment high to comment help degree assigns high weight, away from
The user comment remote from current time assigns low weight, is improved according to comment time weighting to article characteristics.
5. a kind of recommended method to be scored based on user with comment according to claim 1, it is characterised in that: in step 4
In, the calculating article similarity refers to the feature vector for comparing article two-by-two using cosine similarity measure, obtains object
Similarity between product.
6. a kind of recommended method to be scored based on user with comment according to claim 1, it is characterised in that: in step 5
In, the prediction article scoring refers to according to scoring of the similarity calculation user to article between article, obtains user to object
The prediction of product is scored, and detailed process is as follows:
Step 5.1 is ranked up from high to low according to the similarity between article, is obtained and target item most like preceding k
Article；
Step 5.2, the scoring according to user to this preceding k article predict scoring of the user to target item.
7. a kind of recommended method to be scored based on user with comment according to claim 1, it is characterised in that: in step 6
In, the recommendation list that generates refers to the recommendation list initial according to prediction scoring generation of the user to article, and then filtering is used
Scored article at family, generates final recommendation list, detailed process is as follows:
Step 6.1, initial recommendation list generate: being ranked up, obtained just from high to low to the prediction scoring of article according to user
Beginning recommendation list；
Step 6.2, article filtering and sequence: user had given the article of scoring in filtering initial recommendation list, re-started
Sequence；
Step 6.3, consequently recommended list generate: the article sorted lists in read step 6.2, choose the generation of top n article and push away
Recommend list.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910341581.8A CN110110230A (en) | 2019-04-26 | 2019-04-26 | A kind of recommended method to be scored based on user with comment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910341581.8A CN110110230A (en) | 2019-04-26 | 2019-04-26 | A kind of recommended method to be scored based on user with comment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110110230A true CN110110230A (en) | 2019-08-09 |
Family
ID=67486845
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910341581.8A Pending CN110110230A (en) | 2019-04-26 | 2019-04-26 | A kind of recommended method to be scored based on user with comment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110110230A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111460281A (en) * | 2020-02-27 | 2020-07-28 | 浙江口碑网络技术有限公司 | Information push optimization method and device, storage medium and terminal |
| CN112182376A (en) * | 2020-09-28 | 2021-01-05 | 安徽访得信息科技有限公司 | Recommendation engine method of internet advertisement platform capable of real-time and efficient analysis |
| CN113592333A (en) * | 2021-08-07 | 2021-11-02 | 杭州找查科技有限公司 | Internet comprehensive scoring method, system and device and readable storage medium |
| CN113704330A (en) * | 2021-09-23 | 2021-11-26 | 北京网界科技有限公司 | Fair comment system and method thereof |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104715399A (en) * | 2015-04-09 | 2015-06-17 | 苏州大学 | Grading prediction method and grading prediction system |
| CN105095477A (en) * | 2015-08-12 | 2015-11-25 | 华南理工大学 | Recommendation algorithm based on multi-index grading |
| CN106202519A (en) * | 2016-07-22 | 2016-12-07 | 桂林电子科技大学 | A kind of combination user comment content and the item recommendation method of scoring |
| CN106909536A (en) * | 2017-02-05 | 2017-06-30 | 中山大学 | Method is recommended in a kind of scoring based on Heterogeneous Information |
| CN107256494A (en) * | 2017-05-18 | 2017-10-17 | 深圳大学 | A kind of item recommendation method and device |
-
2019
- 2019-04-26 CN CN201910341581.8A patent/CN110110230A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104715399A (en) * | 2015-04-09 | 2015-06-17 | 苏州大学 | Grading prediction method and grading prediction system |
| CN105095477A (en) * | 2015-08-12 | 2015-11-25 | 华南理工大学 | Recommendation algorithm based on multi-index grading |
| CN106202519A (en) * | 2016-07-22 | 2016-12-07 | 桂林电子科技大学 | A kind of combination user comment content and the item recommendation method of scoring |
| CN106909536A (en) * | 2017-02-05 | 2017-06-30 | 中山大学 | Method is recommended in a kind of scoring based on Heterogeneous Information |
| CN107256494A (en) * | 2017-05-18 | 2017-10-17 | 深圳大学 | A kind of item recommendation method and device |
Non-Patent Citations (3)
| Title |
|---|
| LU YICHAO等: "Coevolutionary Recommendation Model: Mutual Learning between Ratings and Reviews", 《THE 2018 WORLD WIDE WEB CONFERENCE》 * |
| 张展鸿: "基于用户评论的推荐系统设计与实现", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 谭云志: "基于用户评分和评论信息的协同推荐框架", 《模式识别与人工智能》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111460281A (en) * | 2020-02-27 | 2020-07-28 | 浙江口碑网络技术有限公司 | Information push optimization method and device, storage medium and terminal |
| CN112182376A (en) * | 2020-09-28 | 2021-01-05 | 安徽访得信息科技有限公司 | Recommendation engine method of internet advertisement platform capable of real-time and efficient analysis |
| CN112182376B (en) * | 2020-09-28 | 2022-07-12 | 安徽多点美好智能科技有限公司 | Recommendation engine method of internet advertisement platform capable of real-time and efficient analysis |
| CN113592333A (en) * | 2021-08-07 | 2021-11-02 | 杭州找查科技有限公司 | Internet comprehensive scoring method, system and device and readable storage medium |
| CN113704330A (en) * | 2021-09-23 | 2021-11-26 | 北京网界科技有限公司 | Fair comment system and method thereof |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110110230A (en) | A kind of recommended method to be scored based on user with comment | |
| CN109815308B (en) | Method and device for determining intention recognition model and method and device for searching intention recognition | |
| CN106528642B (en) | A kind of short text classification method based on TF-IDF feature extractions | |
| CN104391942B (en) | Short essay eigen extended method based on semantic collection of illustrative plates | |
| CN106407236B (en) | A kind of emotion tendency detection method towards comment data | |
| CN110532554A (en) | A kind of Chinese abstraction generating method, system and storage medium | |
| CN101127042A (en) | Sensibility classification method based on language model | |
| CN106708966A (en) | Similarity calculation-based junk comment detection method | |
| CN103365867A (en) | Method and device for emotion analysis of user evaluation | |
| CN103646088A (en) | Product comment fine-grained emotional element extraction method based on CRFs and SVM | |
| CN105630768A (en) | Cascaded conditional random field-based product name recognition method and device | |
| CN104239554A (en) | Cross-domain and cross-category news commentary emotion prediction method | |
| CN111143672B (en) | Knowledge graph-based professional speciality scholars recommendation method | |
| CN105740404A (en) | Label association method and device | |
| CN106339495A (en) | Topic detection method and system based on hierarchical incremental clustering | |
| CN106021329A (en) | A user similarity-based sparse data collaborative filtering recommendation method | |
| CN110598219A (en) | Emotion analysis method for broad-bean-net movie comment | |
| CN107133282B (en) | Improved evaluation object identification method based on bidirectional propagation | |
| CN112015721A (en) | E-commerce platform storage database optimization method based on big data | |
| CN109598586A (en) | A kind of recommended method based on attention model | |
| CN108573041A (en) | Probability matrix based on weighting trusting relationship decomposes recommendation method | |
| CN105512333A (en) | Product comment theme searching method based on emotional tendency | |
| CN110019820A (en) | Main suit and present illness history symptom Timing Coincidence Detection method in a kind of case history | |
| CN112052401A (en) | Recommendation method based on user comments | |
| CN103617245A (en) | Bilingual sentiment classification method and device |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190809 |
|
| RJ01 | Rejection of invention patent application after publication |