WO2012118087A1 - レコメンダシステム、レコメンド方法、及びプログラム - Google Patents
レコメンダシステム、レコメンド方法、及びプログラム Download PDFInfo
- Publication number
- WO2012118087A1 WO2012118087A1 PCT/JP2012/054961 JP2012054961W WO2012118087A1 WO 2012118087 A1 WO2012118087 A1 WO 2012118087A1 JP 2012054961 W JP2012054961 W JP 2012054961W WO 2012118087 A1 WO2012118087 A1 WO 2012118087A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- user
- information
- users
- value
- management unit
- Prior art date
Links
Images
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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2457—Query processing with adaptation to user needs
- G06F16/24578—Query processing with adaptation to user needs using ranking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/335—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0278—Product appraisal
-
- 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
Definitions
- the present invention relates to a recommender system, a recommendation method, and a program.
- a recommender system / platform that provides optimum items in consideration of context information that changes from moment to moment such as individual characteristics, preference information, and circumstances. For this purpose, static information that changes slowly with time, such as preference information of each user, and dynamic information that changes from moment to moment, such as the current position, are handled, and context-aware customized for each user.
- a recommender system is required.
- Patent Document 1 the similarity between items is extracted from an item history purchased by the user in the past, and an evaluation is performed from among items that are highly similar to items already known to be of interest to the user.
- a system for displaying expensive items is described.
- simply displaying items that are rated high on average will provide homogeneous information to all users, and may not work for users with a small number of users or users with a minority preference. There is.
- Context can be static or dynamic.
- An example of a static context that is relatively slow in time is user preference information for items such as books, movies, and news.
- recommended products / items are determined by estimating preference information of active users. The process typically performed is to identify users having a behavior history with high similarity from the behavior history such as the purchase history of the user and display items highly evaluated by those users.
- users who have similar purchase histories are considered to have similar selections, and that an item evaluated by one user is considered highly likely to be highly evaluated by the other user.
- Non-Patent Document 1 describes a system for recommending news based on preference information of a user having a preference tendency similar to that of an active user.
- An example of a recommender system using a dynamic context is a real space application.
- the current position is a representative example of a user context that changes from moment to moment.
- the current position of each user is acquired and managed using GPS or the like, and an item related to the current position is recommended.
- the similarity here is close to the geodetic coordinate distance.
- the real space application is applied to a social networking application that displays other users in the vicinity of an active user, a recommender system that displays popular nearby restaurants, sightseeing spots, and the like.
- Patent Document 2 includes an information registration unit that registers and associates each browsing information with a classification tag that indicates which classification the content corresponds to, and an evaluation by a user who has browsed each browsing information.
- An evaluation acquisition unit to be acquired a user information storage unit for storing evaluations for individual browsing information acquired by the evaluation acquisition unit together with identification information for each user, and a classification tag associated with the individual browsing information
- a user's evaluation is determined by how other users have evaluated the user's evaluation itself.
- the user's own evaluation also differs depending on the user, for example, a user who is highly evaluated by one user i does not always evaluate the same as another user j. Therefore, when the user k (active user) is a minority group compared to the evaluation tendency of a certain user, the value of the recommended item for the user k may be low. That is, when the user evaluation criteria for user k is significantly different from other majority users, a set of users having high similarity to user k is created, and among them, the evaluation of users with high evaluation is weighted. Even if an item is extracted, the value of the item is not necessarily high for the user k.
- an object of the present invention is to perform a highly accurate recommendation even when there are few users with similar preference tendencies.
- a recommender system is a recommender system that provides an information object to a user, and includes an input unit that receives a search request from the user, and a user management unit that manages context information representing the characteristics of the user An index table management unit that manages a plurality of information objects and outputs information objects related to the search request of the user, and similarity between users calculated from a comparison of the context information of the user and other users And a user authority value management unit that calculates an authority value representing the reliability of the user with respect to the other user based on the reliability of the plurality of users with respect to the other user, and the other user with respect to the information object The evaluation value is weighted according to the authority value of the other user A rating calculation unit that calculates an evaluation value of the user for the information object, and an ordering list creation unit that outputs a list of information objects ranked based on the evaluation value of the user. It is provided.
- the block diagram which shows the structure of the recommender system by embodiment of this invention.
- the figure which shows the example of the structure of the object which the recommender system makes object by embodiment of this invention.
- the figure explaining the index table management part in detail by embodiment of this invention.
- FIG. 1 is a block diagram showing a configuration of a recommender system 100 according to an embodiment of the present invention.
- the recommender system 100 includes a user interface unit 101 typified by a browser installed on a computer that operates under program control and a network unit 102 typified by the Internet, which are used by a service recipient (user k, active user). Connected by.
- the recommender system 100 includes an input unit 103 that processes input information from the user k, a user management unit 104 that manages information about the user k, and an index table management unit 105 that indexes and manages objects presented to the user k.
- a user authority value management unit 106 that manages user evaluation information, a rating calculation unit 107 that calculates a rating of an object with respect to the user k, and a set of objects sorted according to the rating information of the object, and ranked.
- An ordered list creation unit 108 for creating a list, and an output unit 109 that processes the processing result so as to be displayed on the user interface unit 101 and outputs the processed result through the network unit 102 are provided.
- the recommender system 100 may be a dedicated or general-purpose computer including a CPU, a memory such as a ROM or a RAM, an external storage device that stores various types of information, an input interface, an output interface, a communication interface, and a bus connecting these. it can.
- the recommender system 100 may be configured by a single computer or may be configured by a plurality of computers connected to each other via a communication line.
- An input unit 103, a user management unit 104, an index table management unit 105, a user authority value management unit 106, a rating calculation unit 107, an ordered list creation unit 108, and an output unit 109 are predetermined programs stored in a ROM or the like by a CPU. It corresponds to the module of the function realized by executing.
- the user interface unit 101 exchanges input / output information with the user k. For example, the user's situation such as a search keyword input and position information of the user k is transmitted to the recommender system 100. Further, the result of sorting the objects related to the search keyword is received from the recommender system 100 and output.
- the network unit 102 is a computer network typified by the Internet that connects the client system (user interface unit 101) of the user k and the recommender system 100.
- the input unit 103 has a function of enabling connection with the user interface unit 101 via the network unit 102, and receives input information to the recommender system 100 from the outside.
- the user management unit 104 manages information on users who access the recommender system 100. Specifically, user profile information, object history and preference information referred to in the past, evaluation information of other users, and the like are managed. Further, based on information received via the input unit 103, a key represented by a keyword is created. When a keyword is directly input from the user, the user information is associated with the keyword. When position information or other indirect information is input, a key closely related to the information is automatically generated, and index table management is performed. Information in a format compatible with the key managed by the unit 105 is generated.
- the index table management unit 105 manages an index table in which objects handled by the recommender system 100 are associated with keywords that are closely related, and outputs a set of objects that are closely related to the keywords in response to keyword input.
- the user authority value management unit 106 outputs authority value information, which is an index for determining the reliability of evaluation of another user for the user k, using the user's action history and the user's mutual evaluation result.
- the rating calculation unit 107 calculates an evaluation value of each object for the user k based on the object information and authority information.
- the ordered list creation unit 108 sorts the objects based on the calculation result of the rating calculation unit 107, and creates a list in which the objects are ordered in the order that suits the user k.
- the output unit 109 processes information so that the user k can correctly browse the ordered list of objects on the user interface unit 101, and outputs the processed list through the network unit 102.
- FIG. 2 is a diagram illustrating an example of the structure of the object 200 targeted by the recommender system 100.
- the object 200 includes a header 201 and a body 202.
- the header 201 includes meta information related to the object, and the body 202 includes actual content information of the object or reference information that can refer to it.
- the content information is information of one movie if, for example, the recommender system 100 is a movie review site. Further, if the recommender system 100 is a music information site, the information is music information.
- Content information is generally a unit of information / data to be handled.
- the header 201 includes general meta information such as an object ID and a category ID for classifying the object, as well as information on an evaluation result of a user who has viewed the object in the past.
- the key information automatically generated by the user management unit 104 can be input as key information of the index table management unit 105, and is information that represents the user's context. For example, it may be a feature keyword of a file that is currently being browsed / edited, or a keyword that expresses information that is closely related to the current position information or time. Examples of information deeply associated with the current location information include information on facilities such as restaurants and sights near the location, and advertising information closely related to the location.
- FIG. 3 is a diagram for explaining the index table management unit 105 in detail.
- the index table management unit 105 is a function for creating an index in advance so that information related to a certain key can be instantaneously referred to from a set of objects managed by the recommender system 100.
- the index table 302 is composed of key / value pairs, and stores information deeply related to the keys as values.
- the key is typically a keyword, and the value is information such as a pointer that can refer to the object set 303 closely related to the key.
- the top object pointer of the object set 303 may be stored, and the objects 304 belonging to the object set 303 may be collected as a linked list so that they can be cross-referenced.
- any value that can refer to a set of objects 304 closely related to a key from a value value may be used.
- the index table management unit 105 searches the corresponding row in the index table 302 in response to the input of the keyword 301, and outputs the object set 303 using the value value of that row.
- the index table 302 can be configured by various methods.
- the index table 302 can be configured by extracting a keyword deeply related to an object by, for example, natural language processing of the object and creating a reverse lookup dictionary thereof.
- FIG. 4 is a diagram illustrating the user authority value management unit 106 in detail.
- the user authority value management unit 106 includes a user reference network management unit 402, a user action history 403, a weight calculation unit 404, and an authority value calculation unit 405, and is based on the user mutual evaluation result 401 and the user action history 403.
- the calculated authority value vector 406 is output.
- the user mutual evaluation result 401 is a user evaluation index which is determined from the evaluation result of each user with respect to other users and who evaluates who and how much. For example, when a user reviews a past review content and evaluates the user highly for the reason that he / she has the same opinion as himself / herself, a high value such as 4 or 5 is assigned in a five-step evaluation. Note that it is not necessary for all users to evaluate each other, and only some users may be evaluated.
- the user reference network management unit 402 uses the user mutual evaluation result 401 to represent the evaluation relationship in a graph. For example, if the evaluation is expressed in binary and the user A evaluates the user B, an edge is created as A-> B, and an edge is not created for the user who does not evaluate. Alternatively, users may be connected with a full mesh, and a user evaluation value may be assigned as the weight.
- the user action history 403 is a past action history of each user used to extract context similarity between users from context information such as user characteristics and preference information.
- the action history can be recorded either directly or indirectly.
- An example of a static context is preference information for an object.
- a direct recording method there is a method of recording what evaluation each user has performed on each object. For example, it is information that the user i has given a 10-level rating of 7 for the object x, and is the information that is input to the rating information of the header 201. It should be noted that not all users need to have an evaluation for an object.
- Indirect recording methods include information based on the number of times an object is referenced based on the click history and the reference time.
- the user's profile (age, sex, address, etc.) and the interest range explicitly indicated by the user himself / herself may be included.
- the user action history as dynamic context information
- positional information such as accumulated data of geodetic coordinate data by GPS.
- transient position information such as where the person has been in the last few hours is strong context information, and the similarity is considered to have a strong correlation with the similarity of necessary information.
- information including both static and dynamic information may be handled.
- These are heterogeneous sets of information, but typically can be expressed as a set of vectors, and the similarity can be defined by the distance between sets.
- the weight calculation unit 404 calculates the similarity between users based on the user action history 403, and outputs the weight used by the authority value calculation unit 405. This weight is information on the importance of other users as seen from each user, and can be typically expressed in a matrix, but is not limited thereto.
- the authority value calculation unit 405 calculates the authority value of each user based on the user reference network management information and the weight information, and outputs authority value information, which is an evaluation value for other users as seen by each user.
- FIG. 5 is a diagram for explaining a weight calculation method of the weight calculation unit 404.
- the weight is calculated from the similarity of the user behavior history, it is necessary to efficiently calculate the similarity with respect to heterogeneous information instead of using a simple set.
- the user action history is once converted into a context expression unit 602 in a compressed form.
- the context expression unit 602 has an information compression effect that expresses user behavior histories representing preferences for highly similar objects as the same except for small differences.
- rating information for an object is user action history.
- the feature of the object x is expressed by a vector v x of an arbitrary dimension (D dimension) and the rating information s x is expressed by a real value
- d (x, y)
- the similarity between the user i and the user j is increased.
- the degree of similarity between the users who have evaluated the same object in the same way as high (low) is increased.
- the following method is an example of a method for realizing the above function.
- the user action history 403 (601 in FIG. 5) holds characteristic information of each object and user preference information acquired by a direct or indirect method.
- feature information is represented by a multidimensional vector v x
- preference information is represented by an integer s x of ⁇ 5 to 5
- (v x , s x ) is held.
- an atomic label value for the object x is calculated using the following hash function f (v).
- C and W are parameters specified in advance by the user, and are a natural number of 2 or more and a real number, respectively.
- R is a uniform random number between 0 and W
- A is a vector having the same dimension as v
- each element is a random number according to an independent standard normal distribution N (0, 1).
- L (v x ) ⁇ f 1 (v x ),..., F B (v x )>, in which B hash functions are independently created and each atomic label is multiplexed, is a label for the object x. And This multiplexed label has a feature that there is a high possibility that objects having similar feature vectors have the same label.
- An extended histogram is a frequency distribution of objects registered in a plurality of bins with different labels.
- it can also have a negative value.
- B L is a bin having a label L
- the frequency h i (L) is the sum of the evaluation values of all objects registered in the bin.
- h i (L) can also be a negative value since s x can have a negative value. If this h i (L) is used, the expanded histogram H i can be expressed by the following equation.
- extended histogram H i can be represented as a point D L dimensional vector when the total number of labels was D L.
- ⁇ e L ⁇ is an orthonormal system
- e L is a unit vector corresponding to the label L.
- the expanded histogram has a large positive frequency for the label corresponding to the preferred object, and a negative negative frequency for the label corresponding to the unfavorable object.
- the frequency of the label corresponding to an object that has been evaluated or hardly evaluated has a value close to zero. Since users having similar preferences have similar extended histograms, the similarity calculation unit 603 can be used to evaluate the similarity.
- the similarity calculation unit 603 evaluates the similarity between users using the extended histogram of each user. For example, an inner product between normalized extended histograms can be used, but the present invention is not necessarily limited thereto.
- the similarity calculation unit 603 outputs a weight 604, which may use the value of the similarity as it is, may be converted into another real number by an appropriate function, and other users viewed from a certain user k It only needs to be an increasing function for the importance of. In other words, the higher the importance, the higher the evaluation value may be.
- scaling is made to be a positive real number in the range of 0 to 1 by adding 1 and dividing by 2.
- the weight 604 is an N ⁇ N matrix, and the value of i row and j column is an importance index for evaluation of the user j as viewed from the user i.
- This processing is basically performed off-line, but information may be updated by performing necessary correction processing on-line.
- FIG. 6 is a diagram illustrating a method in which the authority value calculation unit 405 calculates the authority value of each user viewed from the user k.
- the basic idea is an algorithm (Weighted HITS) in which a weight is added to the HITS algorithm developed for Web page search, and the weight of context similarity is added as a weight. HITS is disclosed in Non-Patent Document 2.
- the authority value of each user expressed by each node of the graph is calculated using the weight 604 output from the weight calculation unit 404 and the inter-user graph G output from the user reference network management unit 402.
- the authority value calculation unit 405 uses this W and G to repeatedly obtain the authority value of the user i according to the following recurrence formula.
- G T is the transpose matrix of the matrix G
- t is an integer t ⁇ 1
- a i (0) 1 for all i.
- ⁇ i indicates a “hub” value
- a i indicates an “authority” value
- ⁇ i indicates a high authority node
- a i Let's say how much is referenced from a hub-like node.
- the importance index W of the node actually pointed is calculated as a weight. That is, the hub user who refers to the user having higher importance for the user k is highly evaluated, and the user who is referred to by the important hub is highly evaluated.
- a certain node 701 points to nodes 702 and 703 as hubs.
- the node 702 has importance and authority values w k, 1 and a k, 1 respectively.
- ⁇ k, i the authority values a k, i are then calculated.
- FIG. 7 is a diagram for explaining a calculation process of rating information for the user k of the object x.
- the rating calculation unit 107 outputs the object rating table 501 output by the index table management unit 105 in response to the key input of the user k and the authority value information of other users for the user k output by the user authority value management unit 106. Based on the user authority value table 502, the rating information for the user k of each object in the object set is calculated.
- the object rating table 501 has one table for all the objects belonging to the object set output by the index table management unit 105, and as shown in FIG. Rating information.
- a set of users registered in the object rating table 501 for the object x is U x
- a rating for the object x of the user i ⁇ U x is r i
- the user authority value table 502 for the user k has evaluation value information of other users defined for the user k
- authority values for the user j are a k , j .
- the weighted rating calculation unit 503 outputs a rating value 504 for the user k as ⁇ r> k for the object x, and there are various specific forms. It can be expressed.
- the following equation represents a weighted average of user rating values using the evaluation weight ⁇ of each user.
- ⁇ (a) is a function uniquely determined for the authority value a, and is generally a monotonically increasing function of a.
- the recommender system 100 may use the user's direct input as a trigger for the operation, or may change dynamic user context information typified by automatically detected position information. Furthermore, a predetermined periodic update may be used as a trigger.
- the above-described extended histogram creation unit is used as the context expression unit 602, and the object history extended histogram creation unit 802 and the GPS history extended histogram creation unit 804 respectively generate an extended histogram as a context representation from the object evaluation history 801 and the GPS history 803. Output.
- GPS data is a three-dimensional vector
- the above hash function that can be defined with respect to the Euclidean distance may be used.
- the cosine similarity Another distance measure such as cos (defined by u ⁇ v /
- a corresponding hash function cannot be defined for some distance defined between the elements of the user behavior history information, for example, an average multiplied by a weight reflecting the evaluation value of the element set is not used without using the hash function.
- representative information of the user action history may be extracted. Here, they are expressed as a point in a vector space of appropriate dimensions (D1 and D2 dimensions).
- the correlation analysis unit 805 calculates a correlation matrix (variance covariance matrix) of context expressions between users.
- a correlation matrix (variance covariance matrix) of context expressions between users.
- it is considered to be a (D1 + D2) -dimensional vector formed from the direct sum of each vector space, and its correlation matrix is created, but each component is weighted appropriately to control the importance of information for each component. May be.
- the principal component analysis unit 806 uses the generated correlation matrix to perform eigenvalue calculation of the matrix, takes out an appropriate number in descending order of the absolute value of the eigenvalue, and considers only the subspace spanned by the eigenvector. Perform dimension compression with. As a result, it is possible to represent each user's context information with a principal component that roughly captures the characteristics of the user context distribution, and the similarity analysis unit 807 has a positional relationship in the subspace represented only by this principal component. User similarity is calculated.
- one method for extracting the principal component is to find a direction in which the variance is large (there is a difference in features for each user).
- the component difference between users can be increased and selected as the principal component.
- the rotation matrix is R ( ⁇ )
- the scale transformation matrix that is multiplied by u in the x axis direction is S (u).
- An arbitrary vector v is expressed by R ( ⁇ ) S (u) R ( ⁇ ) v. Since the output weight generally has a larger value as the degree of similarity is higher, the degree of similarity in this embodiment can be expressed by an inner product between vectors, and this is output as a weight 808 between users.
- p is an integer of 1 or more. Assume that the sum is taken in all context expressions, and in this example, C io and C iG .
- the present invention can be applied to an application for providing products and information optimized for each user in an information providing service based on online shopping or word-of-mouth.
- the present invention is also applicable to a service that senses a dynamic situation such as the current user position and provides information such as news, events, and advertisements according to the situation.
- a recommender system that provides an information object to a user, An input unit for receiving a search request from the user; A user management unit that manages context information representing the characteristics of the user; An index table management unit for managing a plurality of information objects and outputting an information object related to the search request of the user; The reliability of the user with respect to the other user is expressed based on the similarity between the users calculated from the comparison of the context information of the user and the other user and the reliability of the plurality of users with respect to the other user.
- a user authority value management unit for calculating authority values
- a rating calculation unit that calculates the evaluation value of the user for the information object using a value obtained by weighting the evaluation value of the other user for the information object according to the authority value of the other user
- a recommender system comprising: an ordered list creating unit that outputs an ordered list of information objects based on the evaluation value of the user.
- Supplementary note 2 The recommendation according to Supplementary note 1, wherein the search request is a keyword input by the user or a trigger automatically generated by a software program installed in the terminal of the user. Da system.
- the context information is The recommender system according to supplementary note 1, including preference information for each information object and a user action history including the current position of the user.
- the context information is The recommender system according to appendix 1, including past evaluation information for each information object of the user.
- the user authority value management unit comprising a context expression unit that expresses a feature of a specific part in the context information using a probabilistic arithmetic means, and calculates a similarity between users based on the similarity of the specific part. Recommender system.
- Appendix 7 The user authority value management unit The recommender system according to appendix 1, wherein the authority value is calculated based on reliability of a plurality of users with respect to past evaluation information with respect to the information object of the other user and similarity between the users.
- the context expression unit The feature of the specific part is expressed by extracting the principal component from the context information and dimensionally compressing, and the user authority value management unit is based on the positional relationship on the partial space expressed only by the principal component,
- the recommender system according to appendix 6 which calculates similarity between users.
- a recommendation method for providing an information object to a user Receiving a search request from the user; Managing a plurality of information objects and outputting an information object related to the search request of the user;
- the reliability of the user with respect to the other user is expressed based on the similarity between the users calculated from the comparison of the context information of the user and the other user and the reliability of the plurality of users with respect to the other user.
- Calculating an authority value Calculating the user's evaluation value for the information object using a value obtained by weighting the evaluation value of the other user for the information object according to the authority value of the other user; And a step of outputting a list of information objects ordered based on the evaluation value of the user.
- a program that functions as a recommender system that provides information objects to users The computer, An input unit for receiving a search request from the user; A user management unit that manages context information representing the characteristics of the user; An index table management unit for managing a plurality of information objects and outputting an information object related to the search request of the user; The reliability of the user with respect to the other user is expressed based on the similarity between the users calculated from the comparison of the context information of the user and the other user and the reliability of the plurality of users with respect to the other user.
- a user authority value management unit for calculating authority values;
- a rating calculation unit that calculates the evaluation value of the user for the information object using a value obtained by weighting the evaluation value of the other user for the information object according to the authority value of the other user;
- a program that functions as an ordered list creating unit that outputs a list of information objects ordered based on the evaluation value of the user.
- the present invention is suitable for highly accurate recommendation even when there are few users with similar preference tendencies.
- 100 recommender system 101 user interface part, 102 network part, 103 input part, 104 user management part, 105 index table management part, 106 user authority value management part, 107 rating calculation part, 108 ordered list creation part, 109 output Part, 200 object, 201 header, 202 body, 301 keyword, 302 index table, 303 object set, 304 object, 401 user mutual evaluation result, 402 user reference network management part, 403 user action history, 404 weight calculation part, 405 authority Value calculation unit, 406 authority value, 501 object rating table, 502 user authority value table, 03 Weighted rating calculation unit, 504 rating value, 601 user action history, 602 context expression unit, 603 similarity calculation unit, 604 weight, 701, 702, 703, 704, 705, 706 node, 801 object evaluation history, 802 object History extended histogram creation unit, 803 GPS history, 804 GPS history extended histogram creation unit, 805 correlation analysis unit, 806 principal component analysis unit, 807 similarity analysis unit, 808 weight
Abstract
Description
図1は、本発明の実施の形態によるレコメンダシステム100の構成を示すブロック図である。レコメンダシステム100は、サービス受容者(ユーザk、アクティブユーザ)が利用する、プログラム制御により動作するコンピュータ上に実装されたブラウザに代表されるユーザインターフェース部101と、インターネットに代表されるネットワーク部102によって接続される。
(付記1)ユーザに情報オブジェクトを提供するレコメンダシステムであって、
前記ユーザからの検索要求を受信する入力部と、
前記ユーザの特徴を表すコンテキスト情報を管理するユーザ管理部と、
複数の情報オブジェクトを管理し、前記ユーザの前記検索要求に関連する情報オブジェクトを出力するインデックステーブル管理部と、
前記ユーザと他のユーザの前記コンテキスト情報の比較から算出されるユーザ間の類似度と、前記他のユーザに対する複数のユーザの信頼度に基づいて、前記他のユーザに対する前記ユーザの信頼度を表すオーソリティ値を算出するユーザオーソリティ値管理部と、
前記情報オブジェクトに対する前記他のユーザの評価値を前記他のユーザの前記オーソリティ値に応じて重み付けした値を用いて、前記情報オブジェクトに対する前記ユーザの評価値を計算するレーティング計算部と、
前記ユーザの評価値に基づいて序列化された情報オブジェクトのリストを出力する序列化リスト作成部と、を備えたレコメンダシステム。
各々の情報オブジェクトに対する選好情報と、前記ユーザの現在位置を含むユーザ行動履歴を含む、付記1に記載のレコメンダシステム。
前記ユーザの各々の情報オブジェクトに対する過去の評価情報を含む、付記1に記載のレコメンダシステム。
前記コンテキスト情報に基づいて、前記検索要求に関連するキー情報を生成し、
前記インデックステーブル管理部は、
各々の情報オブジェクトを前記キー情報と関連付けて管理し、前記検索要求に関連するキー情報に対応する情報オブジェクトを出力する、付記1に記載のレコメンダシステム。
前記コンテキスト情報の中の特定部分の特徴を確率的な演算手段を用いて表現するコンテキスト表現部を備え、前記特定部分の類似度に基づいて、ユーザ間の類似度を算出する、付記1に記載のレコメンダシステム。
前記他のユーザの情報オブジェクトに対する過去の評価情報に対する、複数のユーザの信頼度と、前記ユーザ間の類似度に基づいて、前記オーソリティ値を算出する、付記1に記載のレコメンダシステム。
前記コンテキスト情報から主成分を抽出して次元圧縮することにより前記特定部分の特徴を表現し、前記ユーザオーソリティ値管理部は、前記主成分のみで表現される部分空間上の位置関係に基づいて、ユーザ間の類似度を算出する、付記6に記載のレコメンダシステム。
前記ユーザからの検索要求を受信する工程と、
複数の情報オブジェクトを管理し、前記ユーザの前記検索要求に関連する情報オブジェクトを出力する工程と、
前記ユーザと他のユーザの前記コンテキスト情報の比較から算出されるユーザ間の類似度と、前記他のユーザに対する複数のユーザの信頼度に基づいて、前記他のユーザに対する前記ユーザの信頼度を表すオーソリティ値を算出する工程と、
前記情報オブジェクトに対する前記他のユーザの評価値を前記他のユーザの前記オーソリティ値に応じて重み付けした値を用いて、前記情報オブジェクトに対する前記ユーザの評価値を計算する工程と、
前記ユーザの評価値に基づいて序列化された情報オブジェクトのリストを出力する工程と、を備えたレコメンド方法。
ユーザに情報オブジェクトを提供するレコメンダシステムとして機能させるプログラムであって、
前記コンピュータを、
前記ユーザからの検索要求を受信する入力部と、
前記ユーザの特徴を表すコンテキスト情報を管理するユーザ管理部と、
複数の情報オブジェクトを管理し、前記ユーザの前記検索要求に関連する情報オブジェクトを出力するインデックステーブル管理部と、
前記ユーザと他のユーザの前記コンテキスト情報の比較から算出されるユーザ間の類似度と、前記他のユーザに対する複数のユーザの信頼度に基づいて、前記他のユーザに対する前記ユーザの信頼度を表すオーソリティ値を算出するユーザオーソリティ値管理部と、
前記情報オブジェクトに対する前記他のユーザの評価値を前記他のユーザの前記オーソリティ値に応じて重み付けした値を用いて、前記情報オブジェクトに対する前記ユーザの評価値を計算するレーティング計算部と、
前記ユーザの評価値に基づいて序列化された情報オブジェクトのリストを出力する序列化リスト作成部、として機能させるプログラム。
Claims (10)
- ユーザに情報オブジェクトを提供するレコメンダシステムであって、
前記ユーザからの検索要求を受信する入力部と、
前記ユーザの特徴を表すコンテキスト情報を管理するユーザ管理部と、
複数の情報オブジェクトを管理し、前記ユーザの前記検索要求に関連する情報オブジェクトを出力するインデックステーブル管理部と、
前記ユーザと他のユーザの前記コンテキスト情報の比較から算出されるユーザ間の類似度と、前記他のユーザに対する複数のユーザの信頼度に基づいて、前記他のユーザに対する前記ユーザの信頼度を表すオーソリティ値を算出するユーザオーソリティ値管理部と、
前記情報オブジェクトに対する前記他のユーザの評価値を前記他のユーザの前記オーソリティ値に応じて重み付けした値を用いて、前記情報オブジェクトに対する前記ユーザの評価値を計算するレーティング計算部と、
前記ユーザの評価値に基づいて序列化された情報オブジェクトのリストを出力する序列化リスト作成部と、を備えたレコメンダシステム。 - 前記検索要求は、前記ユーザによって入力されるキーワード、または前記ユーザの端末に実装されたソフトウェアプログラムによって自動的に生成されるトリガーであることを特徴とする、請求項1に記載のレコメンダシステム。
- 前記コンテキスト情報は、
各々の情報オブジェクトに対する選好情報と、前記ユーザの現在位置を含むユーザ行動履歴を含む、請求項1に記載のレコメンダシステム。 - 前記コンテキスト情報は、
前記ユーザの各々の情報オブジェクトに対する過去の評価情報を含む、請求項1に記載のレコメンダシステム。 - 前記ユーザ管理部は、
前記コンテキスト情報に基づいて、前記検索要求に関連するキー情報を生成し、
前記インデックステーブル管理部は、
各々の情報オブジェクトを前記キー情報と関連付けて管理し、前記検索要求に関連するキー情報に対応する情報オブジェクトを出力する、請求項1に記載のレコメンダシステム。 - 前記ユーザオーソリティ値管理部は、
前記コンテキスト情報の中の特定部分の特徴を確率的な演算手段を用いて表現するコンテキスト表現部を備え、前記特定部分の類似度に基づいて、ユーザ間の類似度を算出する、請求項1に記載のレコメンダシステム。 - 前記ユーザオーソリティ値管理部は、
前記他のユーザの情報オブジェクトに対する過去の評価情報に対する、複数のユーザの信頼度と、前記ユーザ間の類似度に基づいて、前記オーソリティ値を算出する、請求項1に記載のレコメンダシステム。 - 前記コンテキスト表現部は、
前記コンテキスト情報から主成分を抽出して次元圧縮することにより前記特定部分の特徴を表現し、前記ユーザオーソリティ値管理部は、前記主成分のみで表現される部分空間上の位置関係に基づいて、ユーザ間の類似度を算出する、請求項6に記載のレコメンダシステム。 - ユーザに情報オブジェクトを提供するレコメンド方法であって、
前記ユーザからの検索要求を受信する工程と、
複数の情報オブジェクトを管理し、前記ユーザの前記検索要求に関連する情報オブジェクトを出力する工程と、
前記ユーザと他のユーザの前記コンテキスト情報の比較から算出されるユーザ間の類似度と、前記他のユーザに対する複数のユーザの信頼度に基づいて、前記他のユーザに対する前記ユーザの信頼度を表すオーソリティ値を算出する工程と、
前記情報オブジェクトに対する前記他のユーザの評価値を前記他のユーザの前記オーソリティ値に応じて重み付けした値を用いて、前記情報オブジェクトに対する前記ユーザの評価値を計算する工程と、
前記ユーザの評価値に基づいて序列化された情報オブジェクトのリストを出力する工程と、を備えたレコメンド方法。 - コンピュータを、
ユーザに情報オブジェクトを提供するレコメンダシステムとして機能させるプログラムであって、
前記コンピュータを、
前記ユーザからの検索要求を受信する入力部と、
前記ユーザの特徴を表すコンテキスト情報を管理するユーザ管理部と、
複数の情報オブジェクトを管理し、前記ユーザの前記検索要求に関連する情報オブジェクトを出力するインデックステーブル管理部と、
前記ユーザと他のユーザの前記コンテキスト情報の比較から算出されるユーザ間の類似度と、前記他のユーザに対する複数のユーザの信頼度に基づいて、前記他のユーザに対する前記ユーザの信頼度を表すオーソリティ値を算出するユーザオーソリティ値管理部と、
前記情報オブジェクトに対する前記他のユーザの評価値を前記他のユーザの前記オーソリティ値に応じて重み付けした値を用いて、前記情報オブジェクトに対する前記ユーザの評価値を計算するレーティング計算部と、
前記ユーザの評価値に基づいて序列化された情報オブジェクトのリストを出力する序列化リスト作成部、として機能させるプログラム。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2013502372A JP5962926B2 (ja) | 2011-03-03 | 2012-02-28 | レコメンダシステム、レコメンド方法、及びプログラム |
US13/824,771 US9569499B2 (en) | 2011-03-03 | 2012-02-28 | Method and apparatus for recommending content on the internet by evaluating users having similar preference tendencies |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2011046855 | 2011-03-03 | ||
JP2011-046855 | 2011-03-03 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2012118087A1 true WO2012118087A1 (ja) | 2012-09-07 |
Family
ID=46758018
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2012/054961 WO2012118087A1 (ja) | 2011-03-03 | 2012-02-28 | レコメンダシステム、レコメンド方法、及びプログラム |
Country Status (3)
Country | Link |
---|---|
US (1) | US9569499B2 (ja) |
JP (1) | JP5962926B2 (ja) |
WO (1) | WO2012118087A1 (ja) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104428785A (zh) * | 2012-05-04 | 2015-03-18 | 罗文有限公司 | 使用图标的关键字的图标密码设定装置以及图标密码设定方法 |
JP2018124966A (ja) * | 2017-02-01 | 2018-08-09 | 富士通株式会社 | ソーシャル・メディア・アカウントおよびコンテンツの相互に補強するランキング |
WO2018207734A1 (ja) * | 2017-05-11 | 2018-11-15 | 学校法人国際大学 | 情報処理装置、相関モデルのデータ構造、プログラム、情報処理方法および相関モデルのデータ構造の製造方法 |
CN111143684A (zh) * | 2019-12-30 | 2020-05-12 | 腾讯科技(深圳)有限公司 | 基于人工智能的泛化模型的训练方法及装置 |
CN112364151A (zh) * | 2020-10-26 | 2021-02-12 | 西北大学 | 一种基于图、引文和内容的论文混合推荐方法 |
Families Citing this family (27)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10445782B2 (en) * | 2010-12-22 | 2019-10-15 | Facebook, Inc. | Expanded tracking and advertising targeting of social networking users |
JP2014142876A (ja) * | 2013-01-25 | 2014-08-07 | Sony Corp | データ生成装置、コンテンツ再生装置、および記憶媒体 |
CN104516879B (zh) * | 2013-09-26 | 2019-09-13 | Sap欧洲公司 | 用于管理含有具有缺失值的记录的数据库的方法和系统 |
US9336278B2 (en) | 2013-09-30 | 2016-05-10 | Google Inc. | User experience and user flows for third-party application recommendation in cloud storage systems |
US9633081B1 (en) * | 2013-09-30 | 2017-04-25 | Google Inc. | Systems and methods for determining application installation likelihood based on user network characteristics |
US9390141B2 (en) | 2013-09-30 | 2016-07-12 | Google Inc. | Systems and methods for determining application installation likelihood based on probabilistic combination of subordinate methods |
IN2013CH05994A (ja) * | 2013-12-20 | 2015-06-26 | Infosys Ltd | |
WO2015102514A1 (en) * | 2013-12-30 | 2015-07-09 | Odnoklassniki Company Limited | Systems and methods for providing music recommendations |
CN105989146B (zh) * | 2015-02-28 | 2019-07-23 | 阿里巴巴集团控股有限公司 | 对象展示方法及装置 |
CN105989154B (zh) * | 2015-03-03 | 2020-07-14 | 华为技术有限公司 | 相似性度量的方法及设备 |
US11281639B2 (en) | 2015-06-23 | 2022-03-22 | Microsoft Technology Licensing, Llc | Match fix-up to remove matching documents |
US20170004434A1 (en) * | 2015-06-30 | 2017-01-05 | International Business Machines Corporation | Determining Individual Performance Dynamics Using Federated Interaction Graph Analytics |
US20210295344A1 (en) * | 2015-11-09 | 2021-09-23 | Dentsu Inc. | Customer relation management device and method |
CN107798012B (zh) * | 2016-09-05 | 2021-12-14 | 腾讯科技(深圳)有限公司 | 阅读资源评论推送方法和系统 |
KR101871747B1 (ko) * | 2017-04-07 | 2018-06-27 | 주식회사 화성 | 유사성향 기반 사용자-관광상품 추천 방법 및 시스템 |
CN110069717B (zh) * | 2017-07-19 | 2023-11-10 | 阿里巴巴集团控股有限公司 | 一种搜索方法及装置 |
CN109903107A (zh) * | 2017-12-08 | 2019-06-18 | 北京京东尚科信息技术有限公司 | 用于生成信息的方法和装置 |
US11244326B2 (en) * | 2018-03-06 | 2022-02-08 | Verizon Media Inc. | Analytical precursor mining for personalized recommendation |
CN108509560B (zh) * | 2018-03-23 | 2021-04-09 | 广州杰赛科技股份有限公司 | 用户相似度获得方法和装置、设备、存储介质 |
US11010411B2 (en) * | 2019-03-29 | 2021-05-18 | Fuji Xerox Co., Ltd. | System and method automatically sorting ranked items and generating a visual representation of ranked results |
US10867000B2 (en) | 2019-03-31 | 2020-12-15 | Td Ameritrade Ip Company, Inc. | Recommendation system for providing personalized and mixed content on a user interface based on content and user similarity |
US11532025B2 (en) * | 2019-08-12 | 2022-12-20 | International Business Machines Corporation | Deep cognitive constrained filtering for product recommendation |
CN112035738B (zh) * | 2020-08-14 | 2023-09-26 | 北京奇艺世纪科技有限公司 | 一种电子书单推荐方法及装置、电子设备 |
CN112036987B (zh) * | 2020-09-11 | 2024-04-02 | 杭州海康威视数字技术股份有限公司 | 确定推荐商品的方法和装置 |
US20230410159A1 (en) * | 2022-06-15 | 2023-12-21 | At&T Intellectual Property I, L.P. | Method and system for personalizing metaverse object recommendations or reviews |
CN115017417B (zh) * | 2022-08-08 | 2022-11-11 | 杭州实在智能科技有限公司 | 基于标签智能过滤和推荐的政策匹配方法及系统 |
CN117539638B (zh) * | 2024-01-04 | 2024-03-22 | 江西拓荒者科技有限公司 | 一种用于工业大数据平台的数据处理方法及系统 |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2007025925A (ja) * | 2005-07-14 | 2007-02-01 | Fuji Xerox Co Ltd | 関連記述提示システム |
JP2008077386A (ja) * | 2006-09-21 | 2008-04-03 | Sony Corp | 情報処理装置および方法、プログラム並びに記録媒体 |
JP2008165490A (ja) * | 2006-12-28 | 2008-07-17 | Nec Corp | 情報選別装置、情報選別方法、プログラムおよび記録媒体 |
JP2009503751A (ja) * | 2005-08-03 | 2009-01-29 | ウィンク テクノロジ−ズ インコ−ポレイテッド | 関連性判定方法、情報収集方法、オブジェクト組織化方法及び検索システム |
JP2009140042A (ja) * | 2007-12-04 | 2009-06-25 | Sony Corp | 情報処理装置、情報処理方法、およびプログラム |
WO2009097153A1 (en) * | 2008-02-01 | 2009-08-06 | Winston Ronald H | Affinity matching system and method |
Family Cites Families (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6266649B1 (en) | 1998-09-18 | 2001-07-24 | Amazon.Com, Inc. | Collaborative recommendations using item-to-item similarity mappings |
US6968334B2 (en) * | 2001-05-15 | 2005-11-22 | Nokia Corporation | Method and business process to maintain privacy in distributed recommendation systems |
KR101102638B1 (ko) * | 2003-11-13 | 2012-01-04 | 파나소닉 주식회사 | 프로그램 추천장치, 프로그램 추천장치의 프로그램 추천방법 및 기록매체 |
US7698270B2 (en) * | 2004-12-29 | 2010-04-13 | Baynote, Inc. | Method and apparatus for identifying, extracting, capturing, and leveraging expertise and knowledge |
US20070143128A1 (en) * | 2005-12-20 | 2007-06-21 | Tokarev Maxim L | Method and system for providing customized recommendations to users |
US7703030B2 (en) * | 2005-01-11 | 2010-04-20 | Trusted Opinion, Inc. | Method and system for providing customized recommendations to users |
US7590616B2 (en) * | 2006-11-17 | 2009-09-15 | Yahoo! Inc. | Collaborative-filtering contextual model based on explicit and implicit ratings for recommending items |
JP2010536102A (ja) * | 2007-08-08 | 2010-11-25 | ベイノート,インク. | コンテキストに基づくコンテンツレコメンデーションの方法及び装置 |
JP5140449B2 (ja) | 2008-01-31 | 2013-02-06 | 共同印刷株式会社 | 情報提供システム |
JP4678546B2 (ja) * | 2008-09-08 | 2011-04-27 | ソニー株式会社 | 推薦装置および方法、プログラム、並びに記録媒体 |
JP4650541B2 (ja) * | 2008-09-08 | 2011-03-16 | ソニー株式会社 | 推薦装置および方法、プログラム、並びに記録媒体 |
US20100312644A1 (en) * | 2009-06-04 | 2010-12-09 | Microsoft Corporation | Generating recommendations through use of a trusted network |
WO2011052315A1 (ja) * | 2009-10-26 | 2011-05-05 | 日本電気株式会社 | コンテンツ推薦システム、推薦方法及び推薦プログラム |
US20120166377A1 (en) * | 2010-12-27 | 2012-06-28 | Nokia Corporation | Method and apparatus for providing recommendations based on a recommendation model and a context-based rule |
-
2012
- 2012-02-28 JP JP2013502372A patent/JP5962926B2/ja active Active
- 2012-02-28 US US13/824,771 patent/US9569499B2/en active Active
- 2012-02-28 WO PCT/JP2012/054961 patent/WO2012118087A1/ja active Application Filing
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2007025925A (ja) * | 2005-07-14 | 2007-02-01 | Fuji Xerox Co Ltd | 関連記述提示システム |
JP2009503751A (ja) * | 2005-08-03 | 2009-01-29 | ウィンク テクノロジ−ズ インコ−ポレイテッド | 関連性判定方法、情報収集方法、オブジェクト組織化方法及び検索システム |
JP2008077386A (ja) * | 2006-09-21 | 2008-04-03 | Sony Corp | 情報処理装置および方法、プログラム並びに記録媒体 |
JP2008165490A (ja) * | 2006-12-28 | 2008-07-17 | Nec Corp | 情報選別装置、情報選別方法、プログラムおよび記録媒体 |
JP2009140042A (ja) * | 2007-12-04 | 2009-06-25 | Sony Corp | 情報処理装置、情報処理方法、およびプログラム |
WO2009097153A1 (en) * | 2008-02-01 | 2009-08-06 | Winston Ronald H | Affinity matching system and method |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104428785A (zh) * | 2012-05-04 | 2015-03-18 | 罗文有限公司 | 使用图标的关键字的图标密码设定装置以及图标密码设定方法 |
JP2018124966A (ja) * | 2017-02-01 | 2018-08-09 | 富士通株式会社 | ソーシャル・メディア・アカウントおよびコンテンツの相互に補強するランキング |
JP7003481B2 (ja) | 2017-02-01 | 2022-01-20 | 富士通株式会社 | ソーシャル・メディア・アカウントおよびコンテンツの相互に補強するランキング |
WO2018207734A1 (ja) * | 2017-05-11 | 2018-11-15 | 学校法人国際大学 | 情報処理装置、相関モデルのデータ構造、プログラム、情報処理方法および相関モデルのデータ構造の製造方法 |
JPWO2018207734A1 (ja) * | 2017-05-11 | 2020-03-19 | 学校法人国際大学 | 情報処理装置、相関モデルのデータ構造、プログラム、情報処理方法および相関モデルのデータ構造の製造方法 |
CN111143684A (zh) * | 2019-12-30 | 2020-05-12 | 腾讯科技(深圳)有限公司 | 基于人工智能的泛化模型的训练方法及装置 |
CN111143684B (zh) * | 2019-12-30 | 2023-03-21 | 腾讯科技(深圳)有限公司 | 基于人工智能的泛化模型的训练方法及装置 |
CN112364151A (zh) * | 2020-10-26 | 2021-02-12 | 西北大学 | 一种基于图、引文和内容的论文混合推荐方法 |
Also Published As
Publication number | Publication date |
---|---|
US9569499B2 (en) | 2017-02-14 |
US20130185294A1 (en) | 2013-07-18 |
JP5962926B2 (ja) | 2016-08-03 |
JPWO2012118087A1 (ja) | 2014-07-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP5962926B2 (ja) | レコメンダシステム、レコメンド方法、及びプログラム | |
CN107424043B (zh) | 一种产品推荐方法及装置,电子设备 | |
CN105426528B (zh) | 一种商品数据的检索排序方法及系统 | |
TWI636416B (zh) | 內容個人化之多相排序方法和系統 | |
US10210179B2 (en) | Dynamic feature weighting | |
Jiang et al. | Fema: flexible evolutionary multi-faceted analysis for dynamic behavioral pattern discovery | |
CN107644036B (zh) | 一种数据对象推送的方法、装置及系统 | |
WO2018040069A1 (zh) | 信息推荐系统及方法 | |
KR20140091530A (ko) | 소셜 네트워크 피쳐들을 갖는 이름 및 기타 검색 질의의 적합도 | |
CN104216881A (zh) | 一种个性化标签的推荐方法及装置 | |
KR20150023432A (ko) | 사용자 데모그래픽을 추정하는 방법 및 장치 | |
US20130073618A1 (en) | Information Providing System, Information Providing method, Information Providing Device, Program, And Information Storage Medium | |
JP6767342B2 (ja) | 検索装置、検索方法および検索プログラム | |
US11551281B2 (en) | Recommendation engine based on optimized combination of recommendation algorithms | |
CN113850649A (zh) | 一种基于多平台用户数据的定制化推荐方法及推荐系统 | |
US10402414B2 (en) | Scalable system and method for weighted similarity estimation in massive datasets revealed in a streaming fashion | |
Yin et al. | Exploring social activeness and dynamic interest in community-based recommender system | |
CN111159563A (zh) | 用户兴趣点信息的确定方法、装置、设备及存储介质 | |
CN111651678A (zh) | 一种基于知识图谱的个性化推荐方法 | |
CN111429161B (zh) | 特征提取方法、特征提取装置、存储介质及电子设备 | |
Babu et al. | An implementation of the user-based collaborative filtering algorithm | |
CN111654714B (zh) | 信息处理方法、装置、电子设备和存储介质 | |
Zhao et al. | Personalized recommendation by exploring social users’ behaviors | |
TW201243627A (en) | Multi-label text categorization based on fuzzy similarity and k nearest neighbors | |
US20150206220A1 (en) | Recommendation Strategy Portfolios |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 12751811 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13824771 Country of ref document: US |
|
ENP | Entry into the national phase |
Ref document number: 2013502372 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 12751811 Country of ref document: EP Kind code of ref document: A1 |