CN113011950A - Product recommendation method and device - Google Patents
Product recommendation method and device Download PDFInfo
- Publication number
- CN113011950A CN113011950A CN202110341038.5A CN202110341038A CN113011950A CN 113011950 A CN113011950 A CN 113011950A CN 202110341038 A CN202110341038 A CN 202110341038A CN 113011950 A CN113011950 A CN 113011950A
- Authority
- CN
- China
- Prior art keywords
- preference
- user
- time period
- score
- target time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
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
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
Abstract
The application discloses a product recommendation method and device, wherein the method mainly comprises the following steps: acquiring a user preference scoring matrix; constructing a time series model corresponding to each combination based on the user preference scoring matrix; a combination comprising a user and a product; respectively calculating interest weights corresponding to the combinations in the target time period by using the time series models corresponding to the combinations; calculating the prediction preference scores of all combinations in the target time period by utilizing the statistical data of the preference scores in the target time period and the interest weights corresponding to all combinations, supplementing the prediction preference scores into a user preference score matrix, and calculating the similarity of every two users in the target time period based on the user preference score matrix to obtain a plurality of similar user clusters; and respectively recommending the products with the preference scores higher than a preset threshold value of each user in the similar user clusters in a target time period to other users in the similar user clusters aiming at each similar user cluster.
Description
Technical Field
The present application relates to the field of product recommendation technologies, and in particular, to a method and an apparatus for product recommendation.
Background
In recent years, the application of recommendation algorithm-related research is gradually mature, but at the same time, the requirements of users are more personalized, so that more researchers begin to intensively discuss the recommendation algorithm so as to better recommend products to the users.
In the conventional recommendation mode, a collaborative filtering recommendation algorithm is mainly adopted, historical information of a user on products is used as a basis, a user neighbor set similar to a target user is found, and then a plurality of products which are relatively interested by other users in the user neighbor set are recommended to the target user.
However, this method only focuses on the historical information of the user, and the user's interest is influenced by various factors, so the user's interest will change over time, and thus product recommendation is easily performed for the user according to the historical information, which is easy to be uniform. Also, when the available historical information is limited, the accuracy of recommending a product that the user is interested in is relatively low. Therefore, the existing mode can not well promote products meeting the interests and requirements of the users to the users.
Disclosure of Invention
Based on the defects of the prior art, the application provides a product recommendation method and device to solve the problem that the accuracy of the conventional recommendation method is poor.
In order to achieve the above object, the present application provides the following technical solutions:
the application provides a product recommendation method in a first aspect, which includes:
acquiring a user preference scoring matrix; wherein the user preference scoring matrix comprises preference scores of a plurality of users for each product in a plurality of time periods;
constructing a time series model corresponding to each combination based on the user preference scoring matrix; wherein a combination comprises one said user and one said product; the time series model is used for predicting preference scores of the user on the products;
calculating to obtain interest weights corresponding to the combinations in a target time period by using the time series models corresponding to the combinations respectively;
calculating the prediction preference score of each combination in the target time period by using the statistical data of the preference scores in the target time period and the interest weight corresponding to each combination;
completing the predicted preference scores of the combinations in the target time period into the user preference score matrix;
calculating the similarity of every two users in the target time period based on the completed user preference scoring matrix to obtain a plurality of similar user clusters;
and recommending the products with the preference scores of the users in the similar user clusters higher than a preset threshold value in the target time period to other users in the similar user clusters respectively aiming at each similar user cluster.
Optionally, in the above product recommendation method, the calculating, by using the time series model corresponding to each combination, an interest weight corresponding to each combination in a target time period includes:
respectively aiming at each combination, predicting a first preference score, a second preference score and a third preference score corresponding to the combination by utilizing a time series model corresponding to the combination; wherein the first preference score refers to a preference score for a product in the combination by a user in the combination over the target time period; the second preference score refers to a user in the combination having a preference score for a product in the combination over the first N time periods of the target time period; the second preference score refers to a user in the combination who scored preferences for products in the combination over the last N time periods of the target time period;
dividing the sum of the first preference score, the second preference score and the third preference score corresponding to each combination by the preference total score corresponding to each combination to obtain the interest weight corresponding to each combination in the target time period; and the preference sum corresponding to the combination is the preference score sum of all the products of the combined user in the time period from the first N time periods to the last N time periods of the target time period.
Optionally, in the above product recommendation method, the calculating, by using the statistical data of the preference scores in the target time period and the interest weights corresponding to the combinations, the predicted preference score of each combination in the target time period includes:
counting preference scores in a target time period to obtain a plurality of preset type parameters; the preset type parameters comprise the total number of scoring users of each product in the target time period, the number of lowest preference scores of the products which score M positions before ranking, the average score of the preference scores of each product, and the average score of the preference scores of all products;
inputting parameters corresponding to the combinations and interest weights corresponding to the combinations in the preset type parameters into a prediction scoring formula respectively aiming at each combination to obtain prediction preference scores of the combinations at the appointed time; wherein the prediction scoring formula is:
AIM-WR represents a combined user's predicted preference score for a product; u. ofiRepresents the total number of scored users, T, for product imNumber of lowest preference scores, P, representing said products ranked M top in scoreiThe preference score of the product i is represented by an average score, and the preference score of all the products is represented by an average score.
Optionally, in the above product recommendation method, the calculating, based on the complemented user preference score matrix, a similarity between every two users in the target time period to obtain a plurality of similar user clusters includes:
on the basis of the completed user preference score matrix, counting preference scores of the users on the products in the target time period and average preference scores of the users on all the products;
calculating preference scores of the users to the products in the target time period and average scores of the preference scores of the users to all the products in the target time period by using a similarity calculation formula to obtain every two products in the target time periodA similarity of the users; wherein, the similarity calculation formula is as follows:
Sim(u,v)Trepresenting the similarity of the user u and the user v in the target time period; ru,iRepresents a preference score, R, for user u for product i over the target time periodv,iRepresenting a preference score for user v for product i over the target time period;
representing the average score of preference scores of the user u on all products in the target time period;
representing the average score of preference scores of the user v on all products in the target time period; omegau,iRepresenting interest weight corresponding to the combination of the user u and the product i; omegav,iRepresenting interest weight corresponding to the combination of the user v and the product i;
and dividing the users with the similarity larger than a preset threshold into a similar user cluster.
Optionally, in the above product recommendation method, after recommending, for each of the similar user clusters, a product whose preference score of each of the users in the similar user cluster is higher than a preset threshold in the target time period to the other users in the similar user cluster, respectively, the method further includes:
calculating an average absolute error, a root mean square error and an accuracy rate by using each of the prediction preference scores and the real preference scores;
and performing product recommendation evaluation based on the average absolute error, the root mean square error and the accuracy.
A second aspect of the present application provides a product recommendation device, including:
the acquisition unit is used for acquiring a user preference scoring matrix; wherein the user preference scoring matrix comprises preference scores of a plurality of users for each product in a plurality of time periods;
the model building unit is used for building a time series model corresponding to each combination based on the user preference scoring matrix; wherein a combination comprises one said user and one said product; the time series model is used for predicting preference scores of the user on the products;
the weight calculation unit is used for calculating and obtaining interest weights corresponding to the combinations in a target time period by respectively utilizing the time series models corresponding to the combinations;
the scoring unit is used for calculating the prediction preference score of each combination in the target time period by utilizing the statistic data of the preference score in the target time period and the interest weight corresponding to each combination;
a completion unit, configured to complete the predicted preference scores of the combinations in the target time period into the user preference score matrix;
the clustering unit is used for calculating the similarity of every two users in the target time period based on the completed user preference scoring matrix to obtain a plurality of similar user clusters;
and the recommending unit is used for recommending products with preference scores higher than a preset threshold value of each user in the similar user clusters to other users in the similar user clusters in the target time period respectively aiming at each similar user cluster.
Optionally, in the above product recommendation device, the weight calculation unit includes:
the prediction unit is used for predicting a first preference score, a second preference score and a third preference score corresponding to each combination by utilizing the time series model corresponding to the combination; wherein the first preference score refers to a preference score for a product in the combination by a user in the combination over the target time period; the second preference score refers to a user in the combination having a preference score for a product in the combination over the first N time periods of the target time period; the second preference score refers to a user in the combination who scored preferences for products in the combination over the last N time periods of the target time period;
a weight calculating subunit, configured to divide a sum of the first preference score, the second preference score, and the third preference score corresponding to each combination by a preference total score corresponding to each combination, to obtain an interest weight corresponding to each combination in the target time period; and the preference sum corresponding to the combination is the preference score sum of all the products of the combined user in the time period from the first N time periods to the last N time periods of the target time period.
Optionally, in the above product recommendation device, the scoring unit includes:
the first statistic unit is used for carrying out statistics on preference scores in a target time period to obtain a plurality of preset type parameters; the preset type parameters comprise the total number of scoring users of each product in the target time period, the number of lowest preference scores of the products which score M positions before ranking, the average score of the preference scores of each product, and the average score of the preference scores of all products;
a scoring subunit, configured to, for each combination, input a parameter corresponding to the combination and an interest weight corresponding to the combination in the multiple preset type parameters into a prediction scoring formula, so as to obtain a prediction preference score of each combination at the specified time; wherein the prediction scoring formula is:
AIM-WR represents a combined user's predicted preference score for a product; u. ofiRepresents the total number of scored users, T, for product imNumber of lowest preference scores, P, representing said products ranked M top in scoreiThe preference score of the product i is represented by an average score, and the preference score of all the products is represented by an average score.
Optionally, in the above product recommendation device, the clustering unit includes:
the second statistical unit is used for counting preference scores of the users on the products in the target time period and average scores of the users on the preference scores of all the products based on the completed user preference score matrix;
the similarity calculation unit is used for calculating preference scores of the users on the products in the target time period and average scores of the preference scores of the users on all the products in the target time period by using a similarity calculation formula to obtain the similarity of every two users in the target time period; wherein, the similarity calculation formula is as follows:
Sim(u,v)Trepresenting the similarity of the user u and the user v in the target time period; ru,iRepresents a preference score, R, for user u for product i over the target time periodv,iRepresenting a preference score for user v for product i over the target time period;
representing the average score of preference scores of the user u on all products in the target time period;
representing the average score of preference scores of the user v on all products in the target time period; omegau,iRepresenting interest weight corresponding to the combination of the user u and the product i; omegav,iRepresenting interest weight corresponding to the combination of the user v and the product i;
and the dividing unit is used for dividing each user with the similarity larger than a preset threshold into a similar user cluster.
Optionally, in the above product recommendation device, the product recommendation device further includes:
the calculation unit is used for calculating an average absolute error, a root mean square error and an accuracy rate by utilizing the prediction preference scores and the real preference scores;
and the evaluation unit is used for carrying out product recommendation evaluation based on the average absolute error, the root-mean-square error and the accuracy.
According to the product recommendation method, a user preference scoring matrix which comprises preference scores of a plurality of users for products in a plurality of time periods is obtained, then one user and one product are used as a combination, a time series model corresponding to each combination is built based on the user preference scoring matrix, and interest weights corresponding to each combination in a target time period are calculated and obtained by respectively utilizing the time series models corresponding to each combination. And calculating the prediction preference scores of all the combinations in the target time period by utilizing the statistical data of the preference scores in the target time period and the interest weights corresponding to all the combinations, thereby predicting the preference scores by combining a time series model and fully considering the transformation of the user interests along with the time. And the prediction preference scores of all combinations in the target time period are supplemented into the user preference score matrix, so that the supplement of the sparse matrix is realized, and the data volume is ensured. And finally, calculating the similarity of every two users in a target time period based on the completed user preference scoring matrix to obtain a plurality of similar user clusters, and recommending products with preference scores higher than a preset threshold value in the target time period of each user in the similar user clusters to other users in the similar user clusters respectively aiming at each similar user cluster, so that the products meeting the interest and the demand of the users can be accurately recommended to the users.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, it is obvious that the drawings in the following description are only embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
Fig. 1 is a flowchart of a product recommendation method according to an embodiment of the present application;
FIG. 2 is a flowchart of a method for calculating interest weights according to an embodiment of the present disclosure;
FIG. 3 is a flow chart of a method of calculating a prediction preference score according to an embodiment of the present application;
fig. 4 is a flowchart of a method for determining a similar user cluster according to an embodiment of the present application;
FIG. 5 is a schematic structural diagram of a product recommendation device according to another embodiment of the present application;
fig. 6 is a schematic structural diagram of a weight calculation unit according to another embodiment of the present application;
fig. 7 is a schematic structural diagram of a scoring unit according to another embodiment of the present application;
fig. 8 is a schematic structural diagram of a clustering unit according to another embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
In this application, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The embodiment of the application provides a product recommendation method, as shown in fig. 1, specifically including the following steps:
s101, obtaining a user preference scoring matrix, wherein the user preference scoring matrix comprises preference scores of a plurality of users for each product in a plurality of time periods.
Specifically, data such as product information, purchasing behavior information, user feedback information, and the like of each user purchasing a product may be acquired as basic data. And after preprocessing such as emotion, data conversion and normalization is carried out on the basic data, the preference score of the user on the purchased product is obtained. And then dividing preference scores of all users for each product according to time periods, specifically dividing the preference scores of the users for each product in each month, so as to obtain a user preference score matrix. Of course, the division may be performed according to other time periods such as season, week, or day.
S102, constructing a time series model corresponding to each combination based on the user preference scoring matrix, wherein each combination comprises a user and a product.
Wherein the time series model is used to predict the preference scores of the users for products. Specifically, the preference scoring matrix includes preference scores of each user for each product in a plurality of time periods. It is possible to divide the preference of a user for a product at various time periods as an observation sequence and construct a time series model based on each observation sequence. The time series model may specifically adopt a differential Integrated Moving Average Autoregressive model (ARIMA). At this time, the constructed model can be specifically expressed as:
wherein the content of the first and second substances,
a preference score representing a predicted t-period,xt-1To xt-pEach element representing an observation sequence, phi represents a coefficient of an autoregressive model of order p, theta represents a coefficient of a moving average model of order q, epsilon represents an influence factor, and mu is a constant term.
S103, calculating interest weights corresponding to the combinations in the target time period by using the time series models corresponding to the combinations respectively.
It should be noted that the target time period is not particularly limited to a certain time period, and may be set according to a requirement. And the target time period is a time period in which the prediction preference score is required, and the target time period may be one or more.
Wherein, the interest weight can be understood as the weight of the interest of a user in a product to the interest of the user in all products. Therefore, the preference score of a user for a certain product in the target time period can be predicted through the time series model, and the interest weight of the user for the product in the target time period can be obtained based on the ratio of the preference score of the user for the product in the target time period to the preference score of the user for all products in the target time period.
Optionally, as shown in fig. 2, a specific implementation manner of step S103 in the embodiment of the present application specifically includes the following steps:
s201, respectively aiming at each combination, predicting a first preference score, a second preference score and a third preference score corresponding to the combination by using a time series model corresponding to the combination.
Wherein the first preference score refers to a preference score for a user in the combination over a target time period for a product in the combination. The second preference score refers to the user in the combination, scoring the preference of the product in the combination for the first N time periods of the target time period. The second preference score refers to the user in the combination, scoring the preference of the product in the combination for the last N time periods of the target time period.
S202, dividing the sum of the first preference score, the second preference score and the third preference score corresponding to each combination by the preference total score corresponding to each combination to obtain the interest weight corresponding to each combination in the target time period.
And the total preference scores corresponding to one combination are the sum of the preference scores of the users of the combination on all the products in the time periods from the first N time periods to the last N time periods of the target time period. Specifically, in each time period from the first N time periods to the next N time periods of the target time period, the user scores the preference of each product through calculation by a time series model, and then the preference scores are summed to obtain the total preference score.
It should be noted that the interest of the user is constantly changing with time, so in the embodiment of the present application, for the target time period, the interest weight of the user on one product is predicted by considering not only the target time period but also the first N time periods and the last N time periods connected thereto. Specifically, the total sum of the preference scores of the user for one product, which are obtained in the target time period and two time periods adjacent to the target time period, is divided by the total sum of the preference scores of the user for all the products, which are obtained in the target time period and two time periods adjacent to the target time period, so as to obtain the interest weight of the user for one product. That is, it can be specifically expressed as:
wherein the content of the first and second substances,
indicates a first preference score,
Indicating a second preference score,
Indicates a third preference score,
Indicating the combined corresponding preference total score.
And S104, calculating the prediction preference scores of the combinations in the target time period by using the statistical data of the preference scores in the target time period and the interest weights corresponding to the combinations.
Specifically, for each combination, the preference scores generated by the users in each combination and the preference scores obtained by the products in the combination may be counted to obtain the statistical data corresponding to each combination. And then calculating the corresponding statistical data and interest weight of each combination to obtain the prediction preference score of each combination in the target time period.
Specifically, as shown in fig. 3, the specific implementation of step S104 specifically includes the following steps:
s301, counting preference scores in a target time period to obtain a plurality of preset type parameters.
The preset type parameters comprise the total number of the scoring users of each product in the target time period, namely the number of scoring with user preference, the number of lowest scoring preference of the products with M top scoring ranking, the average scoring of the scoring of each product and the average scoring of the scoring of all the products.
S302, inputting a parameter corresponding to one combination of a plurality of preset type parameters and an interest weight corresponding to the combination into a prediction scoring formula respectively aiming at each combination to obtain a prediction preference score of each combination at a specified time.
Wherein, the prediction scoring formula is as follows:
wherein AIM-WR indicates a combined user's predicted preference score for a product; u. ofiRepresents the total number of scored users, T, for product imNumber of lowest preference scores, P, representing products ranked M top in scoreiThe preference score of the product i is represented by an average score, and the preference score of all the products is represented by an average score. Each of the above parameters is obtained by counting data in a target time period.
And S105, complementing the prediction preference scores of all combinations in the target time period into a user preference score matrix.
And S106, calculating the similarity of every two users in the target time period based on the completed user preference scoring matrix to obtain a plurality of similar user clusters.
Specifically, the supplemented user preference score matrix can be used as a basis, the similarity of the users is calculated by using the existing clustering algorithm, and the users are clustered into a plurality of clusters. Wherein, a cluster is a similar user cluster.
However, there is temporal sequencing in view of the variation in user interests over time, and the impossibility for all users to score the same item at the same time. Therefore, when calculating the similarity, the embodiment of the application analyzes the user interest conditions in different time periods, calculates the similarity for the user scores in each time period respectively, and obtains the neighbor sets with higher user similarity in different time periods. Different target time periods can be set, so that similar user clusters with higher user similarity degrees in different time periods can be obtained. Therefore, an implementation manner of step S105 in the embodiment of the present application, as shown in fig. 4, specifically includes the following steps:
s401, based on the completed user preference scoring matrix, counting the preference scores of all users on all products in a target time period and the average preference score of each user on all products.
S402, calculating preference scores of all users to all products in the target time period and average scores of preference scores of all users to all products in the target time period by using a similarity calculation formula to obtain the similarity of every two users in the target time period.
Wherein, the similarity calculation formula is as follows:
Sim(u,v)Trepresenting the similarity of user u and user v within the target time period. Ru,iRepresents a preference score, R, for user u for product i over the target time periodv,iIndicating the user v's preference score for product i over the target time period.
Representing the average score of the user u's preference scores for all products over the target time period.
Indicating that user v scored the preference for all products equally over the target time period. Omegau,iRepresenting the interest weight corresponding to the combination of user u and product i. Omegav,iRepresenting the interest weight corresponding to the combination of user v and product i.
And S403, dividing the users with the similarity larger than a preset threshold into a similar user cluster.
S107, recommending the products with the preference scores higher than the preset threshold value of each user in the similar user clusters in the target time period to other users in the similar user clusters respectively.
Optionally, the top M products with the highest predicted preference scores of each user in the similar user cluster in the target time period may be grouped together into a recommendation list, and then the recommendation list is recommended to each user in the similar user cluster.
Optionally, in another embodiment of the present application, after performing step S107, the method may further include: and calculating the average absolute error, the root mean square error and the accuracy by using the prediction preference scores and the real preference scores, and recommending and evaluating the product based on the average absolute error, the root mean square error and the accuracy.
Wherein the real preference score is a real score of the user on a recommended product after the product is recommended to the user. The accuracy of the product recommendation can be assessed by comparing the predicted data with the actual data.
Specifically, the calculation formula of the average absolute error is as follows:
the root mean square error is:
wherein, PiTo representA predicted preference score for product i; qiA true preference score representing product i; and C represents the number of preference scores of the user on the product.
The accuracy index is the ratio of the recommendation list obtained by the algorithm to the scored items of the user, and the higher the ratio is, the better the proving effect is. The specific calculation formula is as follows:
wherein U represents a user set; ruRepresenting the R products recommended for user u; t isuA collection of products that are of high interest to user u.
It should be noted that the smaller the average absolute error and the root mean square error is, the better the recommendation effect is, and the larger the accuracy is, the higher the recommendation effect is, so that the average absolute error, the root mean square error and the accuracy can be specifically compared with corresponding preset values, thereby realizing the product recommendation evaluation.
According to the product recommendation method provided by the embodiment of the application, a user preference scoring matrix which comprises preference scores of a plurality of users on products in a plurality of time periods is obtained, then one user and one product are used as a combination, a time series model corresponding to each combination is constructed based on the user preference scoring matrix, and interest weights corresponding to the combinations in a target time period are calculated and obtained by respectively utilizing the time series models corresponding to the combinations. And calculating the prediction preference scores of all the combinations in the target time period by utilizing the statistical data of the preference scores in the target time period and the interest weights corresponding to all the combinations, thereby predicting the preference scores by combining a time series model and fully considering the transformation of the user interests along with the time. And the prediction preference scores of all combinations in the target time period are supplemented into the user preference score matrix, so that the supplement of the sparse matrix is realized, and the data volume is ensured. And finally, calculating the similarity of every two users in a target time period based on the completed user preference scoring matrix to obtain a plurality of similar user clusters, and recommending products with preference scores higher than a preset threshold value in the target time period of each user in the similar user clusters to other users in the similar user clusters respectively aiming at each similar user cluster, so that the products meeting the interest and the demand of the users can be accurately recommended to the users.
Another embodiment of the present application provides a product recommendation device, as shown in fig. 5, including the following units:
an obtaining unit 501, configured to obtain a user preference scoring matrix.
The user preference scoring matrix comprises preference scores of various products in various time periods by a plurality of users.
A model building unit 502, configured to build a time series model corresponding to each combination based on the user preference scoring matrix.
Wherein a combination comprises a user and a product. The time series model is used to predict the user's preference scores for the products.
The weight calculating unit 503 is configured to calculate and obtain interest weights corresponding to the combinations in the target time period by using the time series models corresponding to the combinations, respectively.
And a scoring unit 504, configured to calculate a predicted preference score of each combination in the target time period by using the statistical data of the preference scores in the target time period and the interest weight corresponding to each combination.
And a completion unit 505, configured to complete the predicted preference scores of the combinations in the target time period into the user preference score matrix.
And the clustering unit 506 is configured to calculate similarity between every two users in the target time period based on the completed user preference scoring matrix, so as to obtain a plurality of similar user clusters.
And the recommending unit 507 is configured to recommend, to the other users in the similar user cluster, a product with a preference score higher than a preset threshold value for each user in the similar user cluster in a target time period, respectively.
Optionally, in a product recommendation device provided in another embodiment of the present application, the weight calculating unit, as shown in fig. 6, includes the following units:
the prediction unit 601 is configured to predict, for each combination, a first preference score, a second preference score, and a third preference score corresponding to the combination by using the time-series model corresponding to the combination.
Wherein the first preference score refers to a preference score for a user in the combination over a target time period for a product in the combination. The second preference score refers to the user in the combination, scoring the preference of the product in the combination for the first N time periods of the target time period. The second preference score refers to the user in the combination, scoring the preference of the product in the combination for the last N time periods of the target time period.
The weight calculating subunit 602 is configured to divide a sum of the first preference score, the second preference score, and the third preference score corresponding to each combination by the total preference score corresponding to each combination, to obtain an interest weight corresponding to each combination in the target time period.
And combining the corresponding preference sums to obtain the combined preference score sum of all the products in the time periods from the first N time periods to the last N time periods of the target time period.
Optionally, in a product recommendation device provided in another embodiment of the present application, a scoring unit is shown in fig. 7, and includes:
the first statistical unit 701 is configured to count preference scores in a target time period to obtain a plurality of preset type parameters.
The preset type parameters comprise the total number of scoring users of each product in a target time period, the number of lowest preference scores of M products before scoring ranking, the average score of preference scores of each product and the average score of preference scores of all products.
A scoring subunit 702, configured to, for each combination, input a parameter corresponding to the combination and an interest weight corresponding to the combination in the multiple preset type parameters into a prediction scoring formula, so as to obtain a prediction preference score of each combination at a specified time.
Wherein, the prediction scoring formula is as follows:
AIM-WR represents a combined user's predicted preference score for a product; u. ofiRepresents the total number of scored users, T, for product imNumber of lowest preference scores, P, representing products ranked M top in scoreiThe preference score of the product i is represented by an average score, and the preference score of all the products is represented by an average score.
Optionally, in a product recommendation device provided in another embodiment of the present application, a clustering unit is shown in fig. 8, and includes:
a second statistical unit 801, configured to perform statistics on preference scores of each user for each product and an average score of preference scores of each user for all products in a target time period based on the completed user preference score matrix.
The similarity calculation unit 802 is configured to calculate, by using a similarity calculation formula, preference scores of each user for each product in a target time period and an average score of preference scores of each user for all products in the target time period, so as to obtain a similarity between every two users in the target time period.
Wherein, the similarity calculation formula is as follows:
Sim(u,v)Trepresenting the similarity of the user u and the user v in the target time period; ru,iRepresenting a preference score, R, for user u for product i over a target time periodv,iA preference score, representing a user v for a product i within a target time period,
Representing the average score of preference scores of the user u on all products in a target time period;
represents the average score, omega, of the user's v preference scores for all products over a target time periodu,iRepresents the interest weight, omega, corresponding to the combination of user u and product iv,iRepresenting the interest weight corresponding to the combination of user v and product i.
A dividing unit 803, configured to divide each user whose similarity is greater than a preset threshold into a similar user cluster.
Optionally, in the product recommendation device provided in another embodiment of the present application, the product recommendation device may further include:
and the calculating unit is used for calculating the average absolute error, the root mean square error and the accuracy rate by utilizing the prediction preference scores and the real preference scores.
And the evaluation unit is used for carrying out product recommendation evaluation based on the average absolute error, the root mean square error and the accuracy.
Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. A method for recommending products, comprising:
acquiring a user preference scoring matrix; wherein the user preference scoring matrix comprises preference scores of a plurality of users for each product in a plurality of time periods;
constructing a time series model corresponding to each combination based on the user preference scoring matrix; wherein a combination comprises one said user and one said product; the time series model is used for predicting preference scores of the user on the products;
calculating to obtain interest weights corresponding to the combinations in a target time period by using the time series models corresponding to the combinations respectively;
calculating the prediction preference score of each combination in the target time period by using the statistical data of the preference scores in the target time period and the interest weight corresponding to each combination;
completing the predicted preference scores of the combinations in the target time period into the user preference score matrix;
calculating the similarity of every two users in the target time period based on the completed user preference scoring matrix to obtain a plurality of similar user clusters;
and recommending the products with the preference scores of the users in the similar user clusters higher than a preset threshold value in the target time period to other users in the similar user clusters respectively aiming at each similar user cluster.
2. The method according to claim 1, wherein the calculating the interest weight corresponding to each combination in the target time period by using the time series model corresponding to each combination comprises:
respectively aiming at each combination, predicting a first preference score, a second preference score and a third preference score corresponding to the combination by utilizing a time series model corresponding to the combination; wherein the first preference score refers to a preference score for a product in the combination by a user in the combination over the target time period; the second preference score refers to a user in the combination having a preference score for a product in the combination over the first N time periods of the target time period; the second preference score refers to a user in the combination who scored preferences for products in the combination over the last N time periods of the target time period;
dividing the sum of the first preference score, the second preference score and the third preference score corresponding to each combination by the preference total score corresponding to each combination to obtain the interest weight corresponding to each combination in the target time period; and the preference sum corresponding to the combination is the preference score sum of all the products of the combined user in the time period from the first N time periods to the last N time periods of the target time period.
3. The method of claim 1, wherein calculating the predicted preference score for each of the combinations over the target time period using the statistics of the preference scores for the target time period and the interest weights corresponding to each of the combinations comprises:
counting preference scores in a target time period to obtain a plurality of preset type parameters; the preset type parameters comprise the total number of scoring users of each product in the target time period, the number of lowest preference scores of the products which score M positions before ranking, the average score of the preference scores of each product, and the average score of the preference scores of all products;
inputting parameters corresponding to the combinations and interest weights corresponding to the combinations in the preset type parameters into a prediction scoring formula respectively aiming at each combination to obtain prediction preference scores of the combinations at the appointed time; wherein the prediction scoring formula is:
AIM-WR represents a combined user's predicted preference score for a product; u. ofiRepresents the total number of scored users, T, for product imNumber of lowest preference scores, P, representing said products ranked M top in scoreiRepresents the average score of preference scores of products i, and Q represents the preference scores of all productsAnd (5) dividing into average parts.
4. The method of claim 1, wherein the calculating the similarity between every two users in the target time period based on the complemented user preference scoring matrix to obtain a plurality of similar user clusters comprises:
on the basis of the completed user preference score matrix, counting preference scores of the users on the products in the target time period and average preference scores of the users on all the products;
calculating preference scores of the users to the products in the target time period and average scores of the preference scores of the users to all the products in the target time period by using a similarity calculation formula to obtain the similarity of every two users in the target time period; wherein, the similarity calculation formula is as follows:
Sim(u,v)Trepresenting the similarity of the user u and the user v in the target time period; ru,iRepresents a preference score, R, for user u for product i over the target time periodv,iRepresenting a preference score for user v for product i over the target time period;
representing the average score of preference scores of the user u on all products in the target time period;
representing the average score of preference scores of the user v on all products in the target time period; omegau,iRepresenting interest weight corresponding to the combination of the user u and the product i; omegav,iRepresenting interest weight corresponding to the combination of the user v and the product i;
and dividing the users with the similarity larger than a preset threshold into a similar user cluster.
5. The method of claim 1, wherein the recommending, for each of the similar user clusters, a product with a preference score higher than a preset threshold for each of the users in the similar user cluster within the target time period to the rest of the users in the similar user clusters, respectively, further comprises:
calculating an average absolute error, a root mean square error and an accuracy rate by using each of the prediction preference scores and the real preference scores;
and performing product recommendation evaluation based on the average absolute error, the root mean square error and the accuracy.
6. A product recommendation device, comprising:
the acquisition unit is used for acquiring a user preference scoring matrix; wherein the user preference scoring matrix comprises preference scores of a plurality of users for each product in a plurality of time periods;
the model building unit is used for building a time series model corresponding to each combination based on the user preference scoring matrix; wherein a combination comprises one said user and one said product; the time series model is used for predicting preference scores of the user on the products;
the weight calculation unit is used for calculating and obtaining interest weights corresponding to the combinations in a target time period by respectively utilizing the time series models corresponding to the combinations;
the scoring unit is used for calculating the prediction preference score of each combination in the target time period by utilizing the statistic data of the preference score in the target time period and the interest weight corresponding to each combination;
a completion unit, configured to complete the predicted preference scores of the combinations in the target time period into the user preference score matrix;
the clustering unit is used for calculating the similarity of every two users in the target time period based on the completed user preference scoring matrix to obtain a plurality of similar user clusters;
and the recommending unit is used for recommending products with preference scores higher than a preset threshold value of each user in the similar user clusters to other users in the similar user clusters in the target time period respectively aiming at each similar user cluster.
7. The apparatus of claim 6, wherein the weight calculating unit comprises:
the prediction unit is used for predicting a first preference score, a second preference score and a third preference score corresponding to each combination by utilizing the time series model corresponding to the combination; wherein the first preference score refers to a preference score for a product in the combination by a user in the combination over the target time period; the second preference score refers to a user in the combination having a preference score for a product in the combination over the first N time periods of the target time period; the second preference score refers to a user in the combination who scored preferences for products in the combination over the last N time periods of the target time period;
a weight calculating subunit, configured to divide a sum of the first preference score, the second preference score, and the third preference score corresponding to each combination by a preference total score corresponding to each combination, to obtain an interest weight corresponding to each combination in the target time period; and the preference sum corresponding to the combination is the preference score sum of all the products of the combined user in the time period from the first N time periods to the last N time periods of the target time period.
8. The apparatus of claim 6, wherein the scoring unit comprises:
the first statistic unit is used for carrying out statistics on preference scores in a target time period to obtain a plurality of preset type parameters; the preset type parameters comprise the total number of scoring users of each product in the target time period, the number of lowest preference scores of the products which score M positions before ranking, the average score of the preference scores of each product, and the average score of the preference scores of all products;
a scoring subunit, configured to, for each combination, input a parameter corresponding to the combination and an interest weight corresponding to the combination in the multiple preset type parameters into a prediction scoring formula, so as to obtain a prediction preference score of each combination at the specified time; wherein the prediction scoring formula is:
AIM-WR represents a combined user's predicted preference score for a product; u. ofiRepresents the total number of scored users, T, for product imNumber of lowest preference scores, P, representing said products ranked M top in scoreiThe preference score of the product i is represented by an average score, and the preference score of all the products is represented by an average score.
9. The apparatus of claim 6, wherein the clustering unit comprises:
the second statistical unit is used for counting preference scores of the users on the products in the target time period and average scores of the users on the preference scores of all the products based on the completed user preference score matrix;
the similarity calculation unit is used for calculating preference scores of the users on the products in the target time period and average scores of the preference scores of the users on all the products in the target time period by using a similarity calculation formula to obtain the similarity of every two users in the target time period; wherein, the similarity calculation formula is as follows:
Sim(u,v)Trepresenting the similarity of the user u and the user v in the target time period; ru,iIs indicated in the target time periodPreference score, R, for inner user u to product iv,iRepresenting a preference score for user v for product i over the target time period;
representing the average score of preference scores of the user u on all products in the target time period;
representing the average score of preference scores of the user v on all products in the target time period; omegau,iRepresenting interest weight corresponding to the combination of the user u and the product i; omegav,iRepresenting interest weight corresponding to the combination of the user v and the product i;
and the dividing unit is used for dividing each user with the similarity larger than a preset threshold into a similar user cluster.
10. The apparatus of claim 6, further comprising:
the calculation unit is used for calculating an average absolute error, a root mean square error and an accuracy rate by utilizing the prediction preference scores and the real preference scores;
and the evaluation unit is used for carrying out product recommendation evaluation based on the average absolute error, the root-mean-square error and the accuracy.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110341038.5A CN113011950A (en) | 2021-03-30 | 2021-03-30 | Product recommendation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110341038.5A CN113011950A (en) | 2021-03-30 | 2021-03-30 | Product recommendation method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113011950A true CN113011950A (en) | 2021-06-22 |
Family
ID=76409343
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110341038.5A Pending CN113011950A (en) | 2021-03-30 | 2021-03-30 | Product recommendation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113011950A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113610608A (en) * | 2021-08-19 | 2021-11-05 | 创优数字科技(广东)有限公司 | User preference recommendation method and device, electronic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108132964A (en) * | 2017-11-23 | 2018-06-08 | 西北大学 | A kind of collaborative filtering method to be scored based on user item class |
| CN108256093A (en) * | 2018-01-29 | 2018-07-06 | 华南理工大学 | A kind of Collaborative Filtering Recommendation Algorithm based on the more interest of user and interests change |
| CN109543109A (en) * | 2018-11-27 | 2019-03-29 | 山东建筑大学 | A kind of proposed algorithm of time of fusion window setting technique and score in predicting model |
| CN109871479A (en) * | 2019-01-08 | 2019-06-11 | 西北大学 | A kind of collaborative filtering method based on user items class and the reliability that scores |
| CN110427567A (en) * | 2019-07-24 | 2019-11-08 | 东北大学 | A kind of collaborative filtering recommending method based on user preference Similarity-Weighted |
-
2021
- 2021-03-30 CN CN202110341038.5A patent/CN113011950A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108132964A (en) * | 2017-11-23 | 2018-06-08 | 西北大学 | A kind of collaborative filtering method to be scored based on user item class |
| CN108256093A (en) * | 2018-01-29 | 2018-07-06 | 华南理工大学 | A kind of Collaborative Filtering Recommendation Algorithm based on the more interest of user and interests change |
| CN109543109A (en) * | 2018-11-27 | 2019-03-29 | 山东建筑大学 | A kind of proposed algorithm of time of fusion window setting technique and score in predicting model |
| CN109871479A (en) * | 2019-01-08 | 2019-06-11 | 西北大学 | A kind of collaborative filtering method based on user items class and the reliability that scores |
| CN110427567A (en) * | 2019-07-24 | 2019-11-08 | 东北大学 | A kind of collaborative filtering recommending method based on user preference Similarity-Weighted |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113610608A (en) * | 2021-08-19 | 2021-11-05 | 创优数字科技(广东)有限公司 | User preference recommendation method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Caron et al. | Bayesian nonparametric Plackett–Luce models for the analysis of preferences for college degree programmes | |
| CN104462383B (en) | A kind of film based on a variety of behavior feedbacks of user recommends method | |
| CN104573359B (en) | A kind of mass-rent labeled data integration method of task based access control difficulty and mark person's ability | |
| CN103377250B (en) | Top k based on neighborhood recommend method | |
| CN106485562A (en) | A kind of commodity information recommendation method based on user's history behavior and system | |
| CN103136435A (en) | System, method and game platform capable of recommending games in personalization mode | |
| Vahdani et al. | Selecting suppliers using a new fuzzy multiple criteria decision model: the fuzzy balancing and ranking method | |
| CN104281956A (en) | Dynamic recommendation method capable of adapting to user interest changes based on time information | |
| Marley et al. | Best worst scaling: theory and practice | |
| CN108334592A (en) | A kind of personalized recommendation method being combined with collaborative filtering based on content | |
| CN107862022A (en) | Cultural resource commending system | |
| CN102314489A (en) | Method for analyzing opinion leader in network forum | |
| CN112612942B (en) | Social big data-based fund recommendation system and method | |
| CN108648058A (en) | Model sequencing method and device, electronic equipment, storage medium | |
| CN105281925A (en) | Network service user group dividing method and device | |
| CN105354721B (en) | Method and device for identifying machine operation behavior | |
| CN107180088A (en) | News based on Fuzzy C-Means Cluster Algorithm recommends method | |
| CN109657962A (en) | A kind of appraisal procedure and system of the volume assets of brand | |
| CN102999497B (en) | Recommend method and system in a kind of media information position | |
| CN113011950A (en) | Product recommendation method and device | |
| CN112735563A (en) | Recommendation information generation method and device and processor | |
| CN116739669A (en) | System and method for monitoring ocpx advertisements in real time | |
| JP6021223B2 (en) | Matching server | |
| Louviere | Modeling single individuals: the journey from psych lab to the app store | |
| CN116911962B (en) | Article selecting device and method based on data model |
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 |