CN110287402A - A kind of personalized reviews recommended method based on user preference - Google Patents
A kind of personalized reviews recommended method based on user preference Download PDFInfo
- Publication number
- CN110287402A CN110287402A CN201910376763.9A CN201910376763A CN110287402A CN 110287402 A CN110287402 A CN 110287402A CN 201910376763 A CN201910376763 A CN 201910376763A CN 110287402 A CN110287402 A CN 110287402A
- Authority
- CN
- China
- Prior art keywords
- user
- comment
- recommended
- product
- feature
- 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
- 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
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- General Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses the present invention relates to a kind of personalized reviews recommended methods based on user preference, step 1: it obtains user information: providing personalized question and answer to the new user first logged into, the game content of the interested product of its information, especially user and concern is obtained to old user.Step 2: product screening:, may interested product using collaborative filtering recommended user according to the information of user.Step 3: comment is recommended: being directed to the interested product review of user, using the personalized reviews recommended method based on singular value decomposition, predict marking of the user to comment, recommend the comment with high score to user, personalized question and answer are provided to the new user first logged into, its essential information, point of interest and focus are collected to old user.The shortcomings that overcoming lacking individuality in conventional method can show that they are interested to user, have valuable high-quality comment, reach and attract user, and potential consumer is helped to carry out the effect of purchase decision.
Description
Technical field
The present invention relates to network comment screenings to push away correlative technology field, and specially a kind of personalization based on user preference is commented
By recommended method.
Background technique
User can help other potential consumers to understand the product comment of product on e-commerce website or APP
Overall picture helps potential consumer to carry out purchase decision.Such as farm world many farm APP, second farm etc. all use line
The mode received under upper operation, line.In order to attract new user, consolidates the trust of old user, need to provide receipts to potential consumer
Goods shows that is, user receives the product figure of cargo and the evaluation to it.Therefore, how research is sent out from a large amount of product evaluation
Existing valuable comment and to recommend other users particularly significant.
Proposed algorithm can screen the possible interested product of user for the hobby of user according to user information, thus
Recommend the related commentary of the product, the display module that can make to receive is more personalized, reaches guidance user's selection, increases user's letter
Appoint the effect of degree.
Proposed algorithm can be divided into (CF) based on content (CB), based on collaborative filtering and based on mixed method.CB is logical
Content is crossed to calculate the similarity between two products, and CF is the scoring meter of the user by being scored two pieces product
Calculate the similarity of two products.CF is more favourable in nearest recommender system, has also obtained wider application, is most mainstream
Proposed algorithm.Currently, matrix factorisation (MF) becomes most popular one of CF algorithm.Verified, MF surpasses existing big
The algorithm of most technological levels.
Currently, the research of the recommended method and technology of user oriented product review both at home and abroad is not goed deep into still, especially dig
Pick user preferences and interested comment, the automatic assessment and related commentary identification of comment, are organized and recommend aspect matching
Research still belongs to blank.
Comment recommended method based on feature can effectively reflect the quality of comment, predict marking of the consumer to comment
To be recommended.In " Automatically assessing review helpfulness ", " Exploiting social
Context forreview quality prediction ", " Learning to recommend helpful hotel
In the text such as reviews ", researcher attempts to use the marking of text and social characteristic prediction consumer to comment in comment, from
And recommend the comment of those of prediction marking height to consumer.But such methods have ignored the personalized question of consumer, do not examine
Consider the interest of consumer, identical comment may be recommended focus and the different user of point of interest.
Summary of the invention
The personalized reviews recommended method based on user preference that the purpose of the present invention is to provide a kind of, to solve above-mentioned back
The problem of being proposed in scape technology.
To achieve the above object, the invention provides the following technical scheme: a kind of personalized reviews based on user preference push away
Method is recommended, personalized reviews recommended method of this kind based on user preference includes following basic step:
Step 1 obtains user information: providing personalized question and answer to the new user first logged into, obtains its letter to old user
Breath, the especially interested product of user and the game content of concern.The user information includes the area of user, gender with
And a large amount of farm user's history behavioral data.
Step 2, product screening: the user information according to acquired in step 1, using collaborative filtering, recommended user can
It can interested product.
Step 3, comment are recommended: the product recommended according to step 2 is pushed away using the personalized reviews based on singular value decomposition
Method is recommended, marking of the user to comment is predicted, recommends the comment with high score to user.
Preferably, further include process in detail below in the step 2: building user-rating matrix, first building are used
Family-product item rating matrix: being a Matrix C by the data materialization in score information tablem×n, row expression user, list
Aspect mesh, Elements Cu,aScoring of the user u to the project marked as a is represented, score takes the five-grade marking system, minimum value 1, maximum value
It is 5.Numerical value is bigger to illustrate that user is higher to the project degree of recognition.
It finds neighbor user: according to similarity size, determining the highest N number of neighbour of active user's similarity from high to low
It occupies.Similarity is calculated using following formula:
Wherein AP,QThat set includes is user P, the project that Q scored simultaneously,With the meaning phase in formula (2)
Together.
It generates recommended project: taking the corresponding scoring collection of project that family P scored and be combined into H,It is exactly the equal of element sum in H
Value calculates as follows:
Similarly,Meaning is same as above.If the arest neighbors of user P is gathered set S, R after second step calculatesQ,aIt is user Q
Scoring to project a, using following aggregate function prediction user P to the scoring G of project aP,a。
Preferably, in the step 3, the product review of recommended products in step 2 is found, using based on singular value decomposition
Personalized reviews recommended method, predict marking of the user to comment, recommend the comment with high score to user.
Preferably, in the step 3, including following main process: find out the interested product feature of user, and according to
The interested product feature of user is that user constructs user model;According to comment to interested product feature (the i.e. user of user
Model) marking of the description degree prediction user to the comment;According to the prediction score of comment, commented from high to low to user
By recommendation.
Preferably, in the main process of the step 3, original user-eigenmatrix is compressed based on singular value decomposition
It is fitted to a new vector space and by the dot product of two matrixes: the hidden factor matrix of user-and the hidden factor matrix of feature-.It is hidden
The factor is the abstract representation of aobvious factor (product features).
User-eigenmatrix (P) refers to the matrix that all users form the average score of all product features, wherein the
The element P of a row b columna,b, user a is represented to the average score of feature b, and m represents the number of users in set U, and n represents set
The product features quantity that all comments refer to altogether in R.
The hidden factor matrix of user-indicates each user to the interest level of each hidden factor, and the hidden factor matrix of feature-is then
Indicate the semantic relation degree of each feature Yu each hidden factor.The process being entirely fitted is as follows:
Wherein, Vm,kIndicate the hidden factor matrix of user;Sm,kIndicate the hidden factor matrix of feature;It is empty that k represents compressed vector
Between dimension.User a can be by vector v to the interest level of each hidden factora,1,va,2,…va,kIt indicates.For each feature b
For, the semantic relation between b and each hidden factor can be by vector sb,1,sb,2,…,sb,kIt indicates.
Preferably, in the main process of the step 3, first with the relationship between comment and feature, comment is also reflected
It is mapped in the vector space being made of the hidden factor.Relationship between comment and feature can be indicated with following comment-eigenmatrix.
Wherein ci,jUser i is represented to the average score of feature j, N indicates number of reviews in set R, and n represents institute in set R
The product features quantity for thering is comment to refer to altogether.ci,jValue it is higher illustrate feature j appear in comment i in probability it is higher, i.e., two
Correlation degree between person is higher.
Then review map into the vector space being made of the hidden factor:
LN,kIt is comment-hidden factor matrix;lN,kValue more it is high just represent comment l and hidden factor k between relationship it is tighter
It is close.
Preferably, in the main process of the step 3, after comment being also mapped onto hidden factor vector space, handle is only needed
The hidden factor matrix of user-and comment-hidden factor matrix carry out dot product, can obtain each user and beat the prediction of each comment
Point.
Gm,NIt is exactly prediction scoring matrix of the user to comment, the element g in matrixm,NIndicate prediction of the user m to comment N
Marking.Finally, being ranked up according to the prediction marking that user does not read comment to every to comment, and return to wherein highest scoring
User is recommended in the comment of N item.
Compared with prior art, the beneficial effects of the present invention are:
The shortcomings that focusing on user preference, overcoming lacking individuality in conventional method, can show that they are interested to user
, have valuable high-quality comment, reach and attract user, potential consumer is helped to carry out the effect of purchase decision.Recommending to produce
Recommend comment while product, increase the trust of user, more enriches and be conducive to potential consumer and buy.
Detailed description of the invention
Fig. 1 is to be related to a kind of overall procedure schematic diagram of personalized reviews recommended method based on user preference;
Fig. 2 is to be related to a kind of specific flow chart of personalized reviews recommended method based on user preference.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
The present invention provides a kind of technical solution referring to FIG. 1-2: a kind of personalized reviews recommendation side based on user preference
Method, comprising the following specific steps
Step 1 obtains user information: carrying out personalized question and answer to user, provides the new user first logged into personalized
Question and answer collect the game content of the interested product of its information, especially user and concern to old user.The user information
Area including user, gender and a large amount of farm user's history behavioral data.
Step 2, raw product screening: the user information according to acquired in step 1, using collaborative filtering recommended user
It may interested product.Including process in detail below:
User-rating matrix is constructed, is a Matrix C by the data materialization in score information tablem×n, row expression use
Family, list aspect mesh, Elements Cu,aRepresenting scoring of the user u to the project marked as a, score takes the five-grade marking system, minimum value 1,
Maximum value is 5.Numerical value is bigger to illustrate that user is higher to the project degree of recognition.
It finds neighbor user: according to similarity size, determining the Top-N neighbours of active user from high to low.Using
Following formula calculates similarity:
Wherein CP,QThat set includes is user P, the project that Q scored simultaneously,With the meaning phase in formula (2)
Together.
It generates recommended project: taking the corresponding scoring collection of project that family P scored and be combined into H,It is exactly the equal of element sum in H
Value calculates as follows:
Similarly,Meaning is same as above.If the arest neighbors of user P is gathered set S, R after second step calculatesQ,aIt is user Q
Scoring to project a, using following aggregate function prediction user P to the scoring G of project aP,a。
Step 3, comment are recommended: the product recommended according to step 2 is pushed away using the personalized reviews based on singular value decomposition
Method is recommended, marking of the user to comment is predicted, recommends the comment with high score to user.Mainly comprise the processes of find out user sense it is emerging
The product feature of interest, and be that user constructs user model according to the interested product feature of user;Emerging is felt to user according to comment
Marking of the description degree prediction user of the product feature (i.e. user model) of interest to the comment;According to the prediction score of comment,
Comment recommendation is carried out to user from high to low.
Original user-eigenmatrix is compressed to a new vector space based on singular value decomposition and by two matrixes
Dot product fitting: the hidden factor matrix of user-and the hidden factor matrix of feature-.The hidden factor is the abstract table of aobvious factor (product features)
Show.
User-eigenmatrix (P) refers to the matrix that all users form the average score of all product features, wherein the
The element P of a row b columna,b, user a is represented to the average score of feature b, and m represents the number of users in set U, and n represents set
The product features quantity that all comments refer to altogether in R.
The hidden factor matrix of user-indicates each user to the interest level of each hidden factor, and the hidden factor matrix of feature-is then
Indicate the semantic relation degree of each feature Yu each hidden factor.The process being entirely fitted is as follows:
Wherein, Vm,kIndicate the hidden factor matrix of user;Sm,kIndicate the hidden factor matrix of feature;It is empty that k represents compressed vector
Between dimension.User a can be by vector v to the interest level of each hidden factora,1,va,2,…va,kIt indicates.For each feature b
For, the semantic relation between b and each hidden factor can be by vector sb,1,sb,2,…,sb,kIt indicates.
Using the relationship between comment and feature, comment is also mapped onto the vector space being made of the hidden factor.Comment
Relationship between feature can be indicated with following comment-eigenmatrix.
Wherein ci,jUser i is represented to the average score of feature j, N indicates number of reviews in set R, and n represents institute in set R
The product features quantity for thering is comment to refer to altogether.ci,jValue it is higher illustrate feature j appear in comment i in probability it is higher, i.e., two
Correlation degree between person is higher.
Then review map into the vector space being made of the hidden factor:
LN,kIt is comment-hidden factor matrix;lN,kValue more it is high just represent comment l and hidden factor k between relationship it is tighter
It is close.
It, only need to be the hidden factor matrix of user-and comment-hidden factor matrix after comment is also mapped onto hidden factor vector space
Dot product is carried out, each user can be obtained and given a mark to the prediction of each comment.
Gm,NIt is exactly prediction scoring matrix of the user to comment, the element g in matrixm,NIndicate prediction of the user m to comment N
Marking.Finally, being ranked up according to the prediction marking that user does not read comment to every to comment, and return to wherein highest scoring
User is recommended in the comment of N item.
It although an embodiment of the present invention has been shown and described, for the ordinary skill in the art, can be with
A variety of variations, modification, replacement can be carried out to these embodiments without departing from the principles and spirit of the present invention by understanding
And modification, the scope of the present invention is defined by the appended.
Claims (7)
1. a kind of personalized reviews recommended method based on user preference, it is characterised in that: individual character of this kind based on user preference
Changing comment recommended method includes following basic step:
Step 1 obtains user information: personalized question and answer provided to the new user first logged into, its information is obtained to old user,
The especially interested product of user and the game content of concern.The user information includes the area of user, gender and
A large amount of farm user's history behavioral data.
Step 2, product screening: the user information according to acquired in step 1 may be felt using collaborative filtering recommended user
The product of interest.
Step 3, comment are recommended: the product recommended according to step 2, using the personalized reviews recommendation side based on singular value decomposition
Method predicts marking of the user to comment, recommends the comment with high score to user.
2. a kind of personalized reviews recommended method based on user preference according to claim 1, it is characterised in that: described
Further include process in detail below in step 2: building user-rating matrix, first building user-product item score square
Battle array: being a Matrix C by the data materialization in score information tablem×n, row expression user, list aspect mesh, Elements Cu,aIt represents
Scoring of the user u to the project marked as a, score take the five-grade marking system, minimum value 1, maximum value 5.Numerical value is bigger to be illustrated to use
Family is higher to the project degree of recognition.
It finds neighbor user: according to similarity size, determining the highest N number of neighbours of active user's similarity from high to low.It adopts
Similarity is calculated with following formula:
Wherein AP,QThat set includes is user P, the project that Q scored simultaneously,It is identical as the meaning in formula (2).
It generates recommended project: taking the corresponding scoring collection of project that family P scored and be combined into H,It is exactly the mean value of element sum in H, meter
It calculates as follows:
Similarly,Meaning is same as above.If the arest neighbors of user P is gathered set S, R after second step calculatesQ,aIt is user Q to project
The scoring of a, using following aggregate function prediction user P to the scoring G of project aP,a。
3. a kind of personalized reviews recommended method based on user preference according to claim 1, it is characterised in that: described
In step 3, the product review of recommended products in step 2 is found, using the personalized reviews recommendation side based on singular value decomposition
Method predicts marking of the user to comment, recommends the comment with high score to user.
4. a kind of personalized reviews recommended method based on user preference according to claim 1, it is characterised in that: described
In step 3, including following main process: the interested product feature of user is found out, and according to the interested product feature of user
User model is constructed for user;It predicts to use according to description degree of the comment to the interested product feature of user (i.e. user model)
Marking of the family to the comment;According to the prediction score of comment, comment recommendation is carried out to user from high to low.
5. a kind of personalized reviews recommended method based on user preference according to claim 1, it is characterised in that: described
In the main process of step 3, original user-eigenmatrix is compressed to by a new vector space based on singular value decomposition
And it is fitted by the dot product of two matrixes: the hidden factor matrix of user-and the hidden factor matrix of feature-.The hidden factor is that (commodity are special for the aobvious factor
Sign) abstract representation.
User-eigenmatrix (P) refers to all users to the matrix of the average score composition of all product features, wherein a row
The element P of b columna,b, user a is represented to the average score of feature b, and m represents the number of users in set U, and n is represented in set R
The product features quantity that all comments refer to altogether.
The hidden factor matrix of user-indicates each user to the interest level of each hidden factor, and the hidden factor matrix of feature-then indicates
The semantic relation degree of each feature and each hidden factor.The process being entirely fitted is as follows:
Wherein, Vm,kIndicate the hidden factor matrix of user;Sm,kIndicate the hidden factor matrix of feature;K represents compressed vector space
Dimension.User a can be by vector v to the interest level of each hidden factora,1,va,2,…va,kIt indicates.For each feature b
Speech, the semantic relation between b and each hidden factor can be by vector sb,1,sb,2,…,sb,kIt indicates.
6. a kind of personalized reviews recommended method based on user preference according to claim 1, it is characterised in that: described
In the main process of step 3, first with the relationship between comment and feature, comment is also mapped onto and is made of the hidden factor
In vector space.Relationship between comment and feature can be indicated with following comment-eigenmatrix.
Wherein ci,jUser i is represented to the average score of feature j, N indicates number of reviews in set R, and n represents all in set R comment
By the product features quantity referred to altogether.ci,jThe higher probability for illustrating that feature j is appeared in comment i of value it is higher, i.e. the two
Between correlation degree it is higher.
Then review map into the vector space being made of the hidden factor:
LN,kIt is comment-hidden factor matrix;lN,kValue more it is high just represent comment l and hidden factor k between relationship it is closer.
7. a kind of personalized reviews recommended method based on user preference according to claim 1, it is characterised in that: described
In the main process of step 3, after comment being also mapped onto hidden factor vector space, the hidden factor matrix of user-and it need to only comment
Dot product is carried out by-hidden factor matrix, each user can be obtained and given a mark to the prediction of each comment.
Gm,NIt is exactly prediction scoring matrix of the user to comment, the element g in matrixm,NIndicate that user m plays the prediction of comment N
Point.Finally, being ranked up according to the prediction marking that user does not read comment to every to comment, and return to the N of wherein highest scoring
User is recommended in item comment.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910376763.9A CN110287402A (en) | 2019-05-05 | 2019-05-05 | A kind of personalized reviews recommended method based on user preference |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910376763.9A CN110287402A (en) | 2019-05-05 | 2019-05-05 | A kind of personalized reviews recommended method based on user preference |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110287402A true CN110287402A (en) | 2019-09-27 |
Family
ID=68002725
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910376763.9A Pending CN110287402A (en) | 2019-05-05 | 2019-05-05 | A kind of personalized reviews recommended method based on user preference |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110287402A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113741759A (en) * | 2021-11-06 | 2021-12-03 | 腾讯科技(深圳)有限公司 | Comment information display method and device, computer equipment and storage medium |
CN114066572A (en) * | 2021-11-17 | 2022-02-18 | 江南大学 | Cable transaction intelligent recommendation method and system |
-
2019
- 2019-05-05 CN CN201910376763.9A patent/CN110287402A/en active Pending
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113741759A (en) * | 2021-11-06 | 2021-12-03 | 腾讯科技(深圳)有限公司 | Comment information display method and device, computer equipment and storage medium |
CN114066572A (en) * | 2021-11-17 | 2022-02-18 | 江南大学 | Cable transaction intelligent recommendation method and system |
CN114066572B (en) * | 2021-11-17 | 2022-07-12 | 江南大学 | Cable transaction intelligent recommendation method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US20210319346A1 (en) | Personal Taste Assessment Method and System | |
CN101489107B (en) | Collaborative filtering recommendation method based on population attribute keyword vector | |
Ody-Brasier et al. | The price you pay: Price-setting as a response to norm violations in the market for champagne grapes | |
KR102227552B1 (en) | System for providing context awareness algorithm based restaurant sorting personalized service using review category | |
US8930388B2 (en) | System and method for providing orientation into subject areas of digital information for augmented communities | |
CN103365936A (en) | Video recommendation system and method thereof | |
CN107833117A (en) | A kind of Bayes's personalized ordering for considering label information recommends method | |
CN102411754A (en) | Personalized recommendation method based on commodity property entropy | |
CN107391582B (en) | The information recommendation method of user preference similarity is calculated based on context ontology tree | |
CN110399563A (en) | A kind of preference pattern and customized information active recommended method based on user behavior analysis | |
CN104572982B (en) | Personalized recommendation method and system based on problem guiding | |
US11663661B2 (en) | Apparatus and method for training a similarity model used to predict similarity between items | |
CN103559622A (en) | Characteristic-based collaborative filtering recommendation method | |
CN104794207A (en) | Recommendation system based on cooperation and working method of recommendation system | |
CN108563749B (en) | Online education system resource recommendation method based on multi-dimensional information and knowledge network | |
CN106980646A (en) | The method applied based on popularity to the influencing mechanism analysis of user interest and its in proposed algorithm | |
CN108241619A (en) | A kind of recommendation method based on the more interest of user | |
CN110287402A (en) | A kind of personalized reviews recommended method based on user preference | |
CN106951471A (en) | A kind of construction method of the label prediction of the development trend model based on SVM | |
CN109977299A (en) | A kind of proposed algorithm of convergence project temperature and expert's coefficient | |
CN112102029A (en) | Knowledge graph-based long-tail recommendation calculation method | |
CN109885776A (en) | Recommended models can be explained in open source community PR reviewer | |
KR101458588B1 (en) | Expert curation recommendation system and expert recommendation method using thereof by field | |
CN108415928A (en) | A kind of book recommendation method and system based on weighted blend k- nearest neighbor algorithms | |
Li et al. | Identification and evaluation of competitive products based on online user-generated content |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190927 |
|
WD01 | Invention patent application deemed withdrawn after publication |