CN113011942B - Customized product demand collaborative filtering recommendation method based on three-layer neighbor selection framework - Google Patents
Customized product demand collaborative filtering recommendation method based on three-layer neighbor selection framework Download PDFInfo
- Publication number
- CN113011942B CN113011942B CN202110259341.0A CN202110259341A CN113011942B CN 113011942 B CN113011942 B CN 113011942B CN 202110259341 A CN202110259341 A CN 202110259341A CN 113011942 B CN113011942 B CN 113011942B
- Authority
- CN
- China
- Prior art keywords
- neighbor
- user
- layer
- information
- scoring
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 28
- 238000001914 filtration Methods 0.000 title claims abstract description 20
- 238000011156 evaluation Methods 0.000 claims abstract description 34
- 239000011159 matrix material Substances 0.000 claims description 10
- 230000015654 memory Effects 0.000 claims description 10
- 230000007787 long-term memory Effects 0.000 claims description 9
- 230000006403 short-term memory Effects 0.000 claims description 9
- 230000002238 attenuated effect Effects 0.000 claims description 8
- 230000001052 transient effect Effects 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 4
- 230000008859 change Effects 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 3
- 230000002452 interceptive effect Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000013075 data extraction Methods 0.000 description 2
- YOETUEMZNOLGDB-UHFFFAOYSA-N 2-methylpropyl carbonochloridate Chemical compound CC(C)COC(Cl)=O YOETUEMZNOLGDB-UHFFFAOYSA-N 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000005034 decoration Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- 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
- 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
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Development Economics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a custom product demand collaborative filtering recommendation method based on a three-layer neighbor selection framework, which divides a time domain scored by a user into a plurality of time windows, and each window is endowed with a dynamic forgetting factor conforming to the attenuation rule of custom product demand preference information and a suggestion invalidation factor conforming to the suggestion invalidation rule; and combining the two factors to respectively construct a capability evaluation module, a sensitive trust module and a suggestion evaluation module. The three modules are integrated by the three-layer neighbor selection framework, and neighbors with strong prediction capability, credibility and strong suggestion capability are screened out by combining with user scoring information; and finally, predicting item scores of the target user through neighbor score information, and recommending the items possibly interested by the user. The method and the system can capture the preference change of the user, select the item score of the proper neighbor prediction target user and recommend the item, relieve the problem of item recommendation precision reduction caused by data sparseness, and improve the satisfaction degree of the user on the recommended customized product or scheme.
Description
Technical Field
The invention relates to a personalized customized product recommendation technology, in particular to a customized product demand collaborative filtering recommendation method based on a three-layer neighbor selection framework.
Background
In the "internet+" environment, users need to determine their own needs of desired products among a vast and confusing amount of product information. By analyzing explicit (e.g., ratings) or implicit (e.g., web browsing history) user behavior data, the recommendation system can not only shorten the time for a user to find a desired product, but also recommend products that may be of interest to the user to help the user explicitly customize product requirements.
Collaborative Filtering (CF) algorithms are very convenient to process unstructured data for wide application in recommendation systems because field-related feature extraction is not required. CF algorithms can be divided into two classes: (1) Based on the method of item similarity (IBCF), items similar to items that the target user previously liked are recommended. (2) Based on the method of user similarity (UBCF), items of neighbor users that are similar to the target user preferences are recommended. However, the existing method is difficult to process the sparsity problem of the scoring matrix caused by insufficient number of the scoring items of the user and the problem of dynamic change of the preference and the suggestion capability of the user influenced by time factors. Aiming at the problem, the method constructs three neighbor selection modules on the basis of considering the prediction capability, the suggestion capability and the credibility of the user, adopts a three-layer neighbor selection strategy to dynamically evaluate the user capability, and realizes that reliable neighbors can be selected under the data sparse environment.
Disclosure of Invention
The invention aims to provide a custom product demand collaborative filtering recommendation method based on a three-layer neighbor selection framework, which accurately analyzes the preference of a user through the three-layer neighbor selection framework, selects the most suitable neighbor, relieves the problem of reduced recommendation precision caused by data sparseness, better realizes recommendation, and improves the satisfaction degree of the user on recommended items.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
a customized product demand collaborative filtering recommendation method based on a three-layer neighbor selection framework comprises the following steps:
(1) And obtaining user scoring information, and constructing a user scoring matrix, wherein the users comprise neighbor users and target users.
(2) Dividing a time domain scored by a neighbor user into a plurality of time windows, wherein each time window is endowed with a dynamic forgetting factor conforming to the attenuation rule of the customized product demand preference information and a suggestion failure factor conforming to the suggestion failure rule; the preference information attenuation law is that preference information cannot be attenuated in an instantaneous memory stage, rapidly attenuated in a short-term memory stage, and slowly attenuated or even not attenuated in a long-term memory stage. The recommended failure rule is to suggest that no failure occurs in the transient memory phase and that the failure is continued in the short-term memory phase and the long-term memory phase.
(3) And constructing a capability evaluation module and an evaluation formula according to the number of common evaluation items and the asymmetric suggestion capability of the neighbor users and the target users, and calculating the prediction capability of each neighbor user.
(4) And combining the scoring information under each time window and the endowed dynamic forgetting factor, constructing a sensitive trust module to capture the latest preference of each neighbor user and the target user, judging the consistency of the preference, and taking the consistency degree of the preference as the judgment standard of the credibility of each neighbor user.
(5) And constructing a suggestion evaluation module and an evaluation formula according to the number of scoring items for suggestion under each time window and the assigned suggestion failure factor, calculating the effective suggestion information quantity of each neighbor user, and evaluating the suggestion capability of each neighbor user.
(6) Based on three modules, a three-layer neighbor selection strategy is introduced, and a three-layer neighbor selection framework is constructed to obtain a neighbor set with strong prediction capability, credibility and strong suggestion capability;
(7) And (3) predicting the scores of the target users on the various items through the item scores of the various neighbor users in the neighbor set and the capability index values in the step (3-5), and generating a recommendation list for the user to select by using the plurality of items with the highest predicted scores.
Further, in the step (2), the time domain of the score of the neighbor user is divided into a plurality of time windows, wherein the larger the index mark θ value of the time window is, the longer the time period corresponding to the window is from the time below, and the earlier the evaluation time of the score falling into the window is. The dynamic forgetting factors are divided into consistent preference information forgetting factors f con Forgetting factor f of divergence preference information dif The method is specifically expressed as follows:
in the formula ,γi ＝T i 、γ s ＝T i +T s ，T i T is the transient memory stage s T is the short-term memory stage l Is a long-term memory stage. t is t w The length of the time window lambda 1 Is a consistent preference information forgetting rate factor, lambda 2 Is the bifurcating preference information forgetting rate factor, and the larger the numerical value of the bifurcating preference information forgetting rate factor is, the faster the information forgetting rate is.
Further, in step (2), a time window w θ Is marked as θ, which suggests that the failure factor inv satisfies the function:
where ε is the failure rate factor, the speed of suggesting failure is determined.
Further, in the step (3), the evaluation formula is:
in the formula ,|Iu The i represents the total number of items scored by the target user u; i ue The i represents the number of items commonly evaluated by the target user u and the neighbor user e, and the wc (u, e) represents Pearson correlation coefficients of the target user u and the neighbor user e.
Further, the step (4) specifically includes the following sub-steps:
(4.1) the preference information is divided into uniform preference information and divergent preference information, wherein it falls into a time window w θ Is consistent/divergent preference information for individual user scoring of (a)The calculation formula is as follows:
wherein ,Rmax and Rmin Representing the maximum and minimum values of the item score, respectively.Indicating that neighbor user e falls within time window w θ Score on item v, R uv Representing the score of the target user u for the item v. Neighbor user e may be in window w θ In a plurality of common items with the target user u, in each window w θ All the consensus/bifurcation preference part ratings of the following neighbor users e are as follows:
(4.2) the previous scoring preference information decays each time a new score is generated. For a time window w θ Total amount C of coincidence/divergence preference information θ (u,e)/D θ (u, e) is calculated by the formula:
C θ (u,e)＝C θ+1 (u,e)×(1-f con (θ))+c θ (u,e)D θ (u,e)＝D θ+1 (u,e)×(1-f dif (θ))+d θ (u,e)
and (4.3) the sensitive trust module judges the credibility of the neighbor users according to the consistency degree of the preference information among the users. In the time window θ=1, the credibility of the target user u and the neighbor user e is evaluated, and the evaluation formula is as follows:
further, the step (5) specifically comprises:
the maximum number of items I for suggestion in the neighbor user set rmax Can be expressed as
I rmax ＝max|I er |
in the formula ,|Ier The i represents the number of scores that neighbor user e uses for advice, e N '(u), N' (u) represents the neighbor user set.
Each time window w θ Advice information amount m of lower neighbor user e θ Can be expressed as:
in the formula ,time window w is entered in scoring for neighbor user e θ Is a suggested scoring quantity. The amount of suggested information carried by the score may decrease over time. For a time window w θ (II), (III), (V), (; the total amount of advice information for the neighbor users is calculated by the following formula:
M θ ＝M θ+1 ×(1-inv(θ))+m θ
the advice capacity assessment formula for the neighbor user e is:
rec(u,e)＝M 1 /I rmax
in the formula ,M1 For a time window w 1 Total amount of advice information for the lower neighbor user e.
Further, in the step (6), the three-layer neighbor selection policy is to set the number of neighbor selection layers to three, and each layer screens a certain number of neighbors through a key module. Assuming that the number of final neighbor selections is K, the neighbor scaling factor isZeta, the first layer selects front zeta through the ability evaluation module 2 K neighbors with strong prediction capability, the first zeta K neighbors with strong credibility are selected by the second layer through the sensitive trust module, and the first K neighbors with strong suggestion capability are selected by the third layer through the suggestion evaluation module.
Further, in step (7), after the neighbor selection framework selects the neighbor, the score of the target user u on the item v is predicted by the following formula:
wherein ,average value of all item scores for neighbor user e, R ev Scoring on item v for neighbor user e, < +.>And (3) taking an average value of scores of all projects for the target user u, wherein N (u) is a neighbor set obtained by screening in the step (6), per (u, e) is a predictive capability index, tru (u, e) is a credible capability index, and rec (u, e) is a suggested capability index.
Compared with the prior art, the method and the device dynamically analyze the preference of the user through the three-layer neighbor selection framework, select the most suitable neighbor set, predict the item score of the target user by utilizing the neighbor data, relieve the problem of reduced recommendation precision caused by sparse data, better realize recommendation and improve the satisfaction degree of the user on recommended products or schemes.
Drawings
FIG. 1 is a flow chart of a custom product demand collaborative filtering recommendation method based on a three-layer neighbor selection framework.
FIG. 2 is a block diagram of a custom product demand collaborative filtering recommendation system based on a three-layer neighbor selection framework.
Fig. 3 is a process diagram of an implementation of collaborative filtering recommendation for elevator custom product trim requirements.
Detailed Description
The following description of the specific embodiments of the present invention will be more complete and clear with reference to the accompanying drawings, and fig. 1 is a flowchart of a customized product demand collaborative filtering recommendation method based on a three-layer neighbor selection framework, provided by the present invention, including the following steps:
(1) Obtaining user scoring information and constructing a user scoring matrix;
the scoring information of the user can be collected through browsing history, history evaluation data and the like of the user on the website, and is converted into a user scoring matrix containing scoring information of a plurality of users and a plurality of corresponding items, and the scoring matrix is used as data input of the method.
(2) Dividing a time domain into a plurality of time windows, and endowing each window with a dynamic forgetting factor and a suggested failure factor which accord with the attenuation rule and the suggested failure rule of preference information; the concept of decay period of scoring information referring to human brain memory is divided into transient memory phase T i Short-term memory stage T s And long-term memory stage T l The calculation formulas of the time intersection nodes are respectively gamma i ＝T i 、γ s ＝T i +T s . The preference information decay law is that the preference information cannot decay in the transient memory stage, quickly decays in the short-term memory stage, and slowly decays or even does not decay any more in the long-term memory stage. The recommended failure rule is to suggest that no failure occurs in the transient memory phase and that the failure is continuously invalid in the short-term memory phase and the long-term memory phase. In the present embodiment, T i Recommended value range is 24h-72h, T s The recommended value range is 72h-144h. The length of the time window is selected according to actual conditions, and the recommended value range is 12h-72h.
The dynamic forgetting factors are divided into consistent preference information forgetting factors f con Forgetting factor f of divergence preference information dif . Time window w θ The index mark of the current time window is theta, the index mark of the current time window is 1, the index mark value of the current time window is sequentially increased, and the longer the index mark value of the current time window is, the longer the time period corresponding to the window is from the current time, and the earlier the evaluation time of the score falling into the window is. The dynamic forgetting factor for each window satisfies the function:
in the formula ,tw Is the length of the time window lambda 1 Is a consistent preference information forgetting rate factor, lambda 2 Is the bifurcating preference information forgetting rate factor, and the larger the numerical value of the bifurcating preference information forgetting rate factor is, the faster the information forgetting rate is. In the present embodiment, lambda 1 The recommended value of (2) is in the range of 0.07-0.072, lambda 2 The recommended value of (2) is in the range of 0.068-0.07.
The corresponding suggested failure factor inv satisfies the function:
where ε is the failure rate factor, the speed of suggesting failure is determined. In this embodiment, epsilon is recommended to be 0.004-0.006.
(3) Taking the common evaluation quantity and asymmetric suggestion capability of neighbors and target users into consideration, constructing a capability evaluation module and an evaluation formula, wherein the formula is as follows:
in the formula ,|Iu The i represents the total number of items scored by the target user u; i ue The i represents the number of items commonly evaluated by the target user u and the neighbor user e, and the wc (u, e) represents Pearson correlation coefficients of the target user u and the neighbor user e.
(4) Combining the dynamic forgetting factors, constructing a sensitive trust module and a credibility evaluation formula, capturing the latest preference of neighbor users and target users, and judging the consistency of the preference among users;
sensitivity toThe trust module evaluates the credibility of the neighbor users according to the consistency degree of preference information among the users. Preference information may be classified into consistent preference information and divergent preference information. Fall into time window w θ Is a single user scoring of the same/divergent preference information (labeled as and />) The calculation formula is as follows:
wherein ,Rmax and Rmin Representing the maximum and minimum values of the item score, respectively.Indicating that neighbor user e falls within time window w θ Score on item v, R uv Representing the score of the target user u for the item v. User e may be in window w θ In a window w, there are scores of multiple common items with the target user u θ All the consistent/divergent preference part ratings of user e below are:
the previous scoring preference information decays each time a new score is generated. For a time window w θ The total amount of the coincidence/bifurcation preference information (i.e., C θ (u, e) and D θ (u, e)) can be calculated by the following formula:
C θ (u,e)＝C θ+1 (u,e)×(1-f con (θ))+c θ (u,e)D θ (u,e)＝D θ+1 (u,e)×(1-f dif (θ))+d θ (u,e)
in the time window θ=1, the credibility of the target user u and the neighbor user u is evaluated, and the evaluation formula is as follows:
(5) And combining the advice failure factors, constructing an advice evaluation module and an evaluation formula, and calculating the effective advice information amount of the user.
Further, in step (5), the neighbor user sets, the maximum number of items I for suggestion rmax Can be expressed as
I rmax ＝max|I er |
in the formula ,|Ier I is the number of scores that neighbor user e uses for advice, e N '(u), N' (u) represents the neighbor user set.
Each time window w θ Advice information amount m of lower neighbor user e θ Can be expressed as:
in the formula ,time window w is entered in scoring for neighbor user e θ Is a suggested scoring quantity. The amount of suggested information carried by the score may decrease over time. For a time window w θ (II), (III), (V), (; the total amount of advice information for the neighbor users is calculated by the following formula:
M θ ＝M θ+1 ×(1-inv(θ))+m θ
the advice capacity assessment formula for the neighbor user e is:
rec(u,e)＝M 1 /I rmax
in the formula ,M1 For a time window w 1 Total amount of advice information for the lower neighbor user e.
(6) Three layers of neighbor selection strategies are introduced, a neighbor selection framework based on three key modules is constructed, and neighbors with strong prediction capability, credibility and strong suggestion capability are selected.
The three-layer neighbor selection strategy is to set the number of neighbor selection layers as three, and each layer screens a certain number of neighbors through a key module. Assuming that the number of final neighbor selections is K and the neighbor proportion coefficient is ζ, the first layer selects the front ζ through the capability assessment module 2 K neighbors with strong prediction capability, the first zeta K neighbors with strong credibility are selected by the second layer through the sensitive trust module, the first K neighbors with strong suggestion capability are selected by the third layer through the suggestion evaluation module, and the final set is represented as N (u).
(7) Predicting item scores of the target users through the neighbor score data, and generating a recommendation list for the users to select from by using a plurality of items with highest predicted scores.
After the neighbor selection framework selects the neighbors, the scoring of the target user u on the item v is predicted by the following formula:
wherein ,scoring the mean value of the neighbor's items, R ev Scoring on item v for neighbor user e, < +.>The items of the target user u are scored for average value. And finally, generating a recommendation list for the plurality of items with the highest predictive scores to the target user.
As shown in fig. 2, the present invention further designs a collaborative filtering recommendation system based on a three-layer neighbor selection framework, which includes:
(1) Interaction interface: the user can select the product through the interactive interface to express the requirement.
(2) And a data extraction module: and extracting personal information and demand information of the user in the interactive page, retrieving information of similar users and similar items in a user database, and outputting a user scoring matrix.
(3) Capability neighbor selection module: and calculating the ability evaluation score of the similar users according to the predictive ability evaluation formula, and screening out the neighbor set with strong predictive ability of the first layer.
(4) Trusted neighbor selection module: and calculating the credibility of the first layer of neighbors according to the credibility evaluation formula, and screening out the neighbor set with strong credibility of the second layer.
(5) Suggesting neighbor selection module: and calculating the effective advice information quantity of the second-layer neighbors according to the advice capacity assessment formula, and screening out the neighbor set with strong advice capacity of the third layer.
(6) A prediction score calculation module: and predicting the score of each item by the target user through the item score and the capability index value of each neighbor user in the neighbor set.
(7) Item recommendation module: and generating a recommendation list for the plurality of items with the highest predictive scores to the user.
Fig. 3 is a diagram of an implementation of the elevator product decorative demand recommendation according to the present invention.
(1) And the user selects favorite items on the elevator product decoration selection page, and the user demand information and the user information are obtained after the favorite items are processed by the data extraction module. And jointly constructing a user scoring matrix by combining scoring information of the related users and the related items.
(2) And according to the user scoring matrix, the neighbors with strong prediction capability, credibility and strong suggestion capability are screened layer by layer, and the neighbor item scoring data of each layer are reserved through the data registering module, so that the neighbors are asynchronously screened, and the recommendation efficiency is improved.
(3) And calculating the predictive scores of the unscored items of the target user through a predictive score calculation module and a neighbor item score matrix, generating a recommendation list by a plurality of items with highest predictive scores, transmitting the recommendation list to an interactive interface, and recommending the recommendation list to the target user.
The present invention is not limited to the above preferred embodiments, and any person skilled in the art, based on the technical solution of the present invention and the concept of the present invention, can make equivalent substitutions or modifications within the scope of the present invention.
Claims (8)
1. A customized product demand collaborative filtering recommendation method based on a three-layer neighbor selection framework is characterized by comprising the following steps:
(1) Obtaining user scoring information, constructing a user scoring matrix, wherein the user comprises neighbor users and target users;
(2) Dividing a time domain scored by a neighbor user into a plurality of time windows, wherein each time window is endowed with a dynamic forgetting factor conforming to the attenuation rule of the customized product demand preference information and a suggestion failure factor conforming to the suggestion failure rule; the preference information attenuation law is that preference information cannot be attenuated in an instantaneous memory stage, rapidly attenuated in a short-term memory stage and slowly attenuated or even not attenuated in a long-term memory stage; the recommended failure rule is to suggest that no failure occurs in the transient memory stage and that the failure is continuous in the short-term memory stage and the long-term memory stage;
(3) Constructing a capability evaluation module and an evaluation formula according to the number of common evaluation items and asymmetric suggestion capability of the neighbor users and the target users, and calculating the prediction capability of each neighbor user;
(4) Combining the scoring information under each time window and the endowed dynamic forgetting factor, constructing a sensitive trust module to capture the latest preference of each neighbor user and the target user, judging the consistency of the preference, and taking the consistency degree of the preference as the judgment standard of the credibility of each neighbor user;
(5) According to the number of scoring items for advice and the advice failure factor given under each time window, an advice evaluation module and an evaluation formula are constructed, the effective advice information quantity of each neighbor user is calculated, and the advice capacity of each neighbor user is evaluated;
(6) Based on three modules, a three-layer neighbor selection strategy is introduced, and a three-layer neighbor selection framework is constructed to obtain a neighbor set with strong prediction capability, credibility and strong suggestion capability;
(7) And (3) predicting the scores of the target users on the various items through the item scores of the various neighbor users in the neighbor set and the capability index values in the step (3-5), and generating a recommendation list for the user to select by using the plurality of items with the highest predicted scores.
2. The customized product demand collaborative filtering recommendation method based on the three-layer neighbor selection framework of claim 1, which is characterized by comprising the following steps: in the step (2), dividing the time domain of the scoring of the neighbor users into a plurality of time windows, wherein the larger the index mark theta value of the time window is, the longer the time period corresponding to the window is from the time below, and the earlier the scoring evaluation time falling into the window is; the dynamic forgetting factors are divided into consistent preference information forgetting factors f con Forgetting factor f of divergence preference information dif The method is specifically expressed as follows:
in the formula ,γi ＝T i 、γ s ＝T i +T s ，T i T is the transient memory stage s T is the short-term memory stage l T is the long-term memory stage w The length of the time window lambda 1 Is a consistent preference information forgetting rate factor, lambda 2 Is the bifurcating preference information forgetting rate factor, and the larger the numerical value of the bifurcating preference information forgetting rate factor is, the faster the information forgetting rate is.
3. The customized product demand collaboration based on a three-layer neighbor selection framework of claim 1The filtering recommendation method is characterized in that: in step (2), a time window w θ Is marked as θ, which suggests that the failure factor inv satisfies the function:
where ε is the failure rate factor, the speed of suggesting failure is determined.
4. The customized product demand collaborative filtering recommendation method based on the three-layer neighbor selection framework of claim 1, which is characterized by comprising the following steps: in step (3), the evaluation formula is:
in the formula ,|Iu The i represents the total number of items scored by the target user u; i ue The i represents the number of items commonly evaluated by the target user u and the neighbor user e, and the wc (u, e) represents Pearson correlation coefficients of the target user u and the neighbor user e.
5. The customized product demand collaborative filtering recommendation method based on the three-layer neighbor selection framework of claim 1, which is characterized by comprising the following steps: the step (4) specifically comprises the following sub-steps:
(4.1) the preference information is divided into uniform preference information and divergent preference information, wherein it falls into a time window w θ Is consistent/divergent preference information for individual user scoring of (a)The calculation formula is as follows:
wherein ,Rmax and Rmin Representing the maximum and minimum values of the project score respectively,indicating that neighbor user e falls within time window w θ Score on item v, R uv Representing the scoring of item v by target user u, neighbor user e may be in window w θ In a plurality of common items with the target user u, in each window w θ All the consensus/bifurcation preference part ratings of the following neighbor users e are as follows:
(4.2) whenever a new score is generated, the previous scoring preference information decays for time window w θ Total amount C of coincidence/divergence preference information θ (u,e)/D θ (u, e) is calculated by the formula:
C θ (u,e)＝C θ+1 (u,e)×(1-f con (θ))+c θ (u,e)D θ (u,e)＝D θ+1 (u,e)×(1-f dif (θ))+d θ (u,e)
and (4.3) the sensitive trust module judges the credibility of the neighbor users according to the consistency degree of preference information among the users, and evaluates the credibility of the target user u and the neighbor user e under the time window theta=1, wherein the evaluation formula is as follows:
6. the customized product demand collaborative filtering recommendation method based on the three-layer neighbor selection framework of claim 1, which is characterized by comprising the following steps: the step (5) comprises the following steps:
the maximum number of items I for suggestion in the neighbor user set rmax Can be expressed as
I rmax ＝max|I er |
in the formula ,|Ier The I represents the scoring quantity of the neighbor user e for the suggestion, e N '(u), N' (u) represents the neighbor user set;
each time window w θ Advice information amount m of lower neighbor user e θ Can be expressed as:
in the formula ,time window w is entered in scoring for neighbor user e θ The amount of suggested information carried by the score will decrease over time; for a time window w θ (II), (III), (V), (; the total amount of advice information for the neighbor users is calculated by the following formula:
M θ ＝M θ+1 ×(1-inv(θ))+m θ
the advice capacity assessment formula for the neighbor user e is:
rec(u,e)＝M 1 /I rmax
in the formula ,M1 For a time window w 1 Total amount of advice information for the lower neighbor user e.
7. The customized product demand collaborative filtering recommendation method based on the three-layer neighbor selection framework of claim 1, which is characterized by comprising the following steps: in the step (6), the three-layer neighbor selection strategy is to set the number of neighbor selection layers as three, and each layer screens a certain number of neighbors through a key module; assuming that the number of final neighbor selections is K and the neighbor proportion coefficient is ζ, the first layer selects the front ζ through the capability assessment module 2 K neighbors with strong prediction ability, second layer pass sensitivityThe trusted module selects the first ζK neighbors with strong credibility, and the third layer selects the first K neighbors with strong advice capability through the advice evaluation module.
8. The customized product demand collaborative filtering recommendation method based on the three-layer neighbor selection framework of claim 1, which is characterized by comprising the following steps: in the step (7), after the neighbor selection framework selects the neighbor, the score of the target user u on the item v is predicted by the following formula:
wherein ,average value of all item scores for neighbor user e, R ev For scoring of neighbor user e on item v,and (3) taking an average value of scores of all projects for the target user u, wherein N (u) is a neighbor set obtained by screening in the step (6), per (u, e) is a predictive capability index, tru (u, e) is a credible capability index, and rec (u, e) is a suggested capability index.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110259341.0A CN113011942B (en) | 2021-03-10 | 2021-03-10 | Customized product demand collaborative filtering recommendation method based on three-layer neighbor selection framework |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110259341.0A CN113011942B (en) | 2021-03-10 | 2021-03-10 | Customized product demand collaborative filtering recommendation method based on three-layer neighbor selection framework |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113011942A CN113011942A (en) | 2021-06-22 |
| CN113011942B true CN113011942B (en) | 2023-11-03 |
Family
ID=76403877
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110259341.0A Active CN113011942B (en) | 2021-03-10 | 2021-03-10 | Customized product demand collaborative filtering recommendation method based on three-layer neighbor selection framework |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113011942B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20070013371A (en) * | 2005-07-26 | 2007-01-31 | 연세대학교 산학협력단 | Apparatus and method for providing weights to recommendation engines according to situation of user and computer readable medium processing the method |
| KR20090020817A (en) * | 2007-08-24 | 2009-02-27 | 연세대학교 산학협력단 | Collaborative filtering based recommender system and method, neighborgood selection method |
| CN106326390A (en) * | 2016-08-17 | 2017-01-11 | 成都德迈安科技有限公司 | Recommendation method based on collaborative filtering |
| CN108876536A (en) * | 2018-06-15 | 2018-11-23 | 天津大学 | Collaborative filtering recommending method based on arest neighbors information |
| CN110866145A (en) * | 2019-11-06 | 2020-03-06 | 辽宁工程技术大学 | Co-preference assisted deep single-class collaborative filtering recommendation method |
| CN111563787A (en) * | 2020-03-19 | 2020-08-21 | 天津大学 | Recommendation system and method based on user comments and scores |
-
2021
- 2021-03-10 CN CN202110259341.0A patent/CN113011942B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20070013371A (en) * | 2005-07-26 | 2007-01-31 | 연세대학교 산학협력단 | Apparatus and method for providing weights to recommendation engines according to situation of user and computer readable medium processing the method |
| KR20090020817A (en) * | 2007-08-24 | 2009-02-27 | 연세대학교 산학협력단 | Collaborative filtering based recommender system and method, neighborgood selection method |
| CN106326390A (en) * | 2016-08-17 | 2017-01-11 | 成都德迈安科技有限公司 | Recommendation method based on collaborative filtering |
| CN108876536A (en) * | 2018-06-15 | 2018-11-23 | 天津大学 | Collaborative filtering recommending method based on arest neighbors information |
| CN110866145A (en) * | 2019-11-06 | 2020-03-06 | 辽宁工程技术大学 | Co-preference assisted deep single-class collaborative filtering recommendation method |
| CN111563787A (en) * | 2020-03-19 | 2020-08-21 | 天津大学 | Recommendation system and method based on user comments and scores |
Non-Patent Citations (2)
| Title |
|---|
| 融合改进加权Slope One的协同过滤算法;王家华;谈国新;张文元;王阳;杨观赐;;微电子学与计算机(第04期);41-46 * |
| 融合用户兴趣和评分差异的协同过滤推荐算法;陆航;师智斌;刘忠宝;计算机工程与应用(第007期);24-29 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113011942A (en) | 2021-06-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Da Silva et al. | An evolutionary approach for combining results of recommender systems techniques based on collaborative filtering | |
| CN110162706B (en) | Personalized recommendation method and system based on interactive data clustering | |
| CN108256093B (en) | Collaborative filtering recommendation algorithm based on multiple interests and interest changes of users | |
| Atanassov et al. | Intuitionistic fuzzy interpretations of multi-criteria multi-person and multi-measurement tool decision making | |
| CN109543109B (en) | Recommendation algorithm integrating time window technology and scoring prediction model | |
| US8650149B2 (en) | Portable inferred interest and expertise profiles | |
| Cui et al. | A novel context-aware recommendation algorithm with two-level SVD in social networks | |
| Anand et al. | Folksonomy-based fuzzy user profiling for improved recommendations | |
| Ismail et al. | Collaborative filtering-based recommendation of online social voting | |
| CN107038184B (en) | A kind of news recommended method based on layering latent variable model | |
| WO2020078818A1 (en) | Adapting prediction models | |
| CN116308685B (en) | Product recommendation method and system based on aspect emotion prediction and collaborative filtering | |
| Linda et al. | Effective context-aware recommendations based on context weighting using genetic algorithm and alleviating data sparsity | |
| CN114298783A (en) | Commodity recommendation method and system based on matrix decomposition and fusion of user social information | |
| CN113011942B (en) | Customized product demand collaborative filtering recommendation method based on three-layer neighbor selection framework | |
| Yan et al. | A stochastic dominance based approach to consumer-oriented Kansei evaluation with multiple priorities | |
| Lee et al. | Optimal statistical design of a multivariate EWMA chart based on ARL and MRL | |
| CN109670914B (en) | Product recommendation method based on time dynamic characteristics | |
| CN111445280A (en) | Model generation method, restaurant ranking method, system, device and medium | |
| Amin et al. | A maximum discrimination DEA method for ranking association rules in data mining | |
| CN110825967B (en) | Recommendation list re-ranking method for improving diversity of recommendation system | |
| Palomares et al. | Multi-view data approaches in recommender systems: an overview | |
| Yadav et al. | An efficient collaborative recommender system for textbooks using silhouette index and K-means clustering technique | |
| Lee et al. | Abnormal usage sequence detection for identification of user needs via recurrent neural network semantic variational autoencoder | |
| Zhou | Design of a Hybrid Recommendation Algorithm based on Multi-objective Collaborative Filtering for Massive Cloud Data |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |