CN110362755A - A kind of recommended method of the hybrid algorithm based on article collaborative filtering and correlation rule - Google Patents

Abstract

The invention belongs to proposed algorithm technical fields, disclose a kind of recommended method of hybrid algorithm based on article collaborative filtering and correlation rule.By the historical record data collection for constructing user behavior, obtain the correlation rule of article, user-article rating matrix is constructed again, to calculate article similarity, then article-contents attribute matrix is constructed, the similarity based on goods attribute is calculated, article scoring is finally predicted according to similarity, and generate corresponding recommendation list.Using, based on the recommended method of hybrid algorithm, can effectively solve the data sparsity problem encountered in Collaborative Filtering Recommendation Algorithm, and improve the calculation method of similarity in the present invention, to improve recommendation quality.

Description

A Recommendation Method Based on a Hybrid Algorithm of Item Collaborative Filtering and Association Rules

technical field

The invention belongs to the technical field of recommendation algorithms, and in particular relates to a recommendation method based on a hybrid algorithm of item collaborative filtering and association rules.

Background technique

With the rapid development of the Internet industry, the transmission speed of information is getting faster and faster, and the types and quantities of information are increasing. In the face of increasingly complex information, it becomes difficult for users to find the content they like, and the time to search is also getting longer and longer. In order to solve this problem, personalized recommendation technology, namely recommendation algorithm, has been developed rapidly. The recommendation algorithm obtains user preferences by analyzing the user's historical behavior data, so as to actively recommend content that may be of interest to the user.

Traditional recommendation algorithms include: content-based recommendation algorithms, collaborative filtering-based recommendation algorithms, and social network-based recommendation algorithms. At present, the most widely used recommendation algorithm is based on collaborative filtering, which can be divided into user-based collaborative filtering algorithm and item-based collaborative filtering algorithm. The item-based collaborative filtering algorithm is currently the most widely used recommendation algorithm in the industry. It can recommend some items similar to the items they liked before to users. It calculates the similarity between items by analyzing the user's behavior records, which means that item A and item B have great similarity because most users who like item A also like item B. But the similarity calculated by this method does not mean that the two items are similar in attributes. In addition, the algorithm based on collaborative filtering is very dependent on historical data, and when the data is sparse, the quality of its recommendation is seriously degraded. The present invention improves the recommendation quality by using the item rating predicted by the association rule and calculating the item attribute similarity, combined with the item-based collaborative filtering algorithm.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a recommendation method based on a hybrid algorithm of item collaborative filtering and association rules, solve the data sparsity problem encountered in collaborative filtering recommendation algorithm, and improve the calculation method of similarity, thereby improving recommendation quality. In order to achieve the above object, the present invention provides the following technical solutions: a recommendation method based on a hybrid algorithm of item collaborative filtering and association rules, comprising the following steps:

S1. Build a data set of historical records to obtain association rules of items;

S2, construct a user-item rating matrix, and calculate the item similarity;

S3. Construct an item-content attribute matrix, and calculate the similarity based on item attributes;

S4. Predict item ratings according to similarity.

Further, the step S1 includes the following specific steps:

S1-1. Construct a historical record data set according to the user's behavior records each time, and each row of the data set is recorded as a set of items for which positive feedback (for example, purchases) occurred in each behavior record of the user;

S1-2, find out the association rule of the second-order frequent itemsets according to the set minimum support and minimum confidence.

Further, the step S2 includes the following specific steps:

S2-1. Construct a user-item scoring matrix according to the user's historical records and scoring records. Each row of the matrix is the evaluation score of different users on different items, and each column of the matrix is a different item;

S2-2. Use the association rule in step 1 to predict the rating of an item that has had positive feedback with a certain user, and fill the rating into the above-mentioned user-item rating matrix to obtain a new user-item rating matrix;

S2-3. Calculate the similarity between different items by using the cosine similarity formula according to the user-item rating matrix after filling the rating.

Further, the step S3 includes the following specific steps:

S3-1. Construct all item-content attribute matrix, whether each row of the matrix has the content attribute of different items;

S3-2. According to the item-content attribute matrix, the cosine similarity is used to calculate the similarity between different items.

Further, the step S4 includes the following specific steps:

S4-1. The similarity between item i and item j is:

sim(i,j) represents the similarity between item i and item j, and the value range is [0,1]. α represents the weight, and the value range is [0.1, 0.9].

N(i) represents the number of users who like item i, and N(i)∩N(j) represents the number of users who like item i and item j at the same time.

V(i) represents the attribute vector of item i;

S4-2, the calculation formula of item i prediction score is:

P _ui represents user u’s predicted rating for item i. S(i,K) is the set of K items most similar to item i. r _uj represents the real rating of user u to item j;

S4-3. Predict the rating of a user for an item according to a predictive rating formula.

Compared with the prior art, the beneficial effects of the present invention are:

1. Solve the problem that the sparsity of the data using the user-item rating matrix alone leads to the decline of the recommendation quality.

2. Improve the accuracy of similarity prediction item rating and improve the quality of recommendation.

Description of drawings

1 is a schematic flowchart of a recommendation method based on a hybrid algorithm of item collaborative filtering and association rules in the present invention;

FIG. 2 is a comparison of the MAE of the predicted score between the hybrid recommendation algorithm of the present invention and the traditional collaborative filtering algorithm.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

In addition, the technical solutions between the various embodiments of the present invention can be combined with each other, but must be based on the realization by those of ordinary skill in the art. When the combination of technical solutions is contradictory or cannot be achieved, it should be considered that the combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

As shown in Figure 1, S1 constructs a historical record data set and obtains the association rules of items, which includes the following steps:

1. Construct a historical data set, each row of data in the data set is a list of items for which the user has positive feedback: [ItemID1, ItemID2, ..., ItemIDn].

2. Find out the second-order frequent itemsets association rules that satisfy the minimum support degree of 0.2 and the minimum confidence degree of 0.6 through the Apriori association rule algorithm. The Apriori association rule algorithm mainly includes two steps: first, find all the frequent itemsets in the data set, the frequency of these itemsets should be greater than or equal to the minimum support degree, and then find the association rules that satisfy the minimum confidence degree according to the frequent itemsets.

The formula for calculating support:

Support(X→Y) represents the support degree, that is, the probability of item set X and item set Y appearing at the same time. N(X, Y) represents the number of times the item set X and item Y appear at the same time, and N _all represents the total number of historical records.

Confidence calculation formula:

Conf(X→Y) represents the confidence, that is, the probability that item set Y appears at the same time when item set X appears.

3. Traverse the historical record data set, first find the first-order frequent itemsets, then find the second-order frequent itemsets, and find the association rules with confidence greater than or equal to the minimum confidence according to the second-order frequent itemsets. The association rule finally found is a dataset consisting of (ItemIDx, ItemIDy, Confidence), where ItemIDx represents the ID of item x, and Confidence represents the confidence of this set of association rules.

S2, constructing a user-item rating matrix and calculating item similarity, including the following specific steps:

1. According to the user's historical behavior records, construct the user-item rating matrix R:

m represents the number of users, and n represents the number of items. r _ui (1≤u≤m, 1≤i≤n) represents the rating of item i by user u. If user u does not rate item i, then r _ui is empty.

2. Use the association law dataset from step 1 to predict scores. Traverse each row of data in the user-item rating matrix. For each user, if there is an item in the association law dataset for which positive feedback has occurred and the associated item corresponding to the item has not had positive feedback with the user, predict this The user's rating for this associated item. The predicted scoring formula is:

P _ui is the predicted rating of item i by user u, S(u) represents the set of items that user u has rated, and the table conf _ji

is the confidence of item j on item i, and r _uj is the rating of user u on item j.

3. Populate the user-item rating matrix with the predicted ratings. Use the predicted rating above for each item that the user has not rated yet as the user's rating for that item.

4. Calculate item similarity using the populated user-item rating matrix. The similarity calculation formula is:

sim(i,j) is the similarity between item i and item j, and the value range is [0,1]. N(i) represents the number of users who like item i, and N(i)∩N(j) represents the number of users who like item i and item j at the same time.

5. Calculate the similarity of each item with other items, and finally obtain the item similarity matrix S:

s _ij (1≤i, j≤n) represents the similarity between item i and item j.

S3. Construct an item-content attribute matrix, and calculate the similarity based on item attributes. The specific steps are as follows:

1. Analyze each item and construct the item-content attribute matrix W:

n represents the number of items, and l represents the number of item attributes. w _ij (1≤i≤n, 1≤j≤l) represents the jth attribute of item i, if item i has this attribute, then w _ij =1, otherwise w _ij =0.

2. Obtain the similarity between items by comparing the attributes of each item. The similarity calculation formula is:

sim _c (i,j) represents the similarity between item i and item j calculated through content attributes, and the value range is [0,1]. V(i) represents the attribute vector of item i.

3. Calculate the similarity of each item with other items, and finally obtain the item similarity matrix S _c :

S _cij (1≤i,j≤n) represents the similarity between item i and item j calculated based on content attributes.

S4, predicting the item rating according to the similarity, including the following specific steps:

1. Combine the similarity sim(i,j) and the similarity sim _c (i,j) calculated based on the content attribute to obtain the improved item similarity, the formula is: sim _new (i,j)=αsim(i, j)+(1-α)sim _c (i,j).

2. Use the improved item similarity to predict and score the target item i of user u. The prediction scoring formula is:

P _ui represents user u’s predicted rating for item i. S(i,K) is the set of K items most similar to item i. r _uj represents the true rating of item j by user u.

The predicted item score is obtained through S1-4, and the corresponding recommendation list is produced. For the target user, traverse all item sets, select the item sets that have not yet experienced positive feedback, and sort them according to the predicted scores from large to small to generate a list to be recommended. Finally, select the top n items in the list to be recommended to generate a recommendation list.

The recommendation quality is measured by calculating the mean absolute deviation (MAE) between the predicted score and the actual score. The smaller the mean absolute error value, the higher the prediction accuracy. The calculation formula is:

MAE _u represents the MAE of user _u 's predicted rating of Nu items, Nu is the number of items to be predicted and rated, P _ui is the predicted rating of user _{u for item i, and r ui is} the actual rating of user u for item i.

MAE represents the MAE of all users, and M is the number of all users.

Example 1

In the embodiment of the present invention, movielens-latest-small is used as a data set to verify the effect of the method of the present invention. This dataset contains a movie dataset and a rating dataset, the movie dataset contains 9742 movie data, and the rating dataset is 100836 ratings data generated by 610 users for these 9742 movies.

In this embodiment, the scoring data set is first divided into a training data set and a test data set according to a ratio of 7:1, and then several different algorithms are trained on the training data set respectively, and α=0.9 is taken when calculating the similarity, Finally, these algorithms are tested on the test dataset and the resulting MAEs are compared. The results are shown in Figure 2. The abscissa represents the number K of the most similar items selected when calculating the similarity, the ordinate represents the MAE of the predicted score and the actual score, itemCF represents the MAE result generated by the original item-based collaborative filtering algorithm, itemCF -content represents the MAE result generated by the improved similarity method, rule-itemCF represents the MAE result generated by the item-based collaborative filtering algorithm after filling the user-rating matrix with the predicted score using the association rule, rule-itemCF-content is this The MAE generated by the proposed recommendation algorithm.

From the experimental results, it can be seen that the MAE value obtained by the recommendation algorithm proposed in this paper is the smallest, and the recommendation quality is improved to a certain extent. And when the most similar 5 items are used to predict the rating, the resulting MAE value is the smallest.

The above descriptions are only the preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above-mentioned embodiments, but any equivalent modifications or changes made by those of ordinary skill in the art based on the contents disclosed in the present invention should be included in the within the scope of protection described in the claims.

Claims (5)

1. a recommendation method based on the hybrid algorithm of article collaborative filtering and association rules, is characterized in that: comprise the steps: S1. Build a data set of historical records to obtain association rules of items; S2, construct a user-item rating matrix, and calculate the item similarity; S3. Construct an item-content attribute matrix, and calculate the similarity based on item attributes; S4. Predict item ratings according to similarity. 2. A recommendation method based on a hybrid algorithm of item collaborative filtering and association rules according to claim 1, wherein the step S1 comprises the following specific steps: S1-1. Construct a historical record data set according to the user's behavior records each time, and each row of the data set is recorded as a set of items that have positive feedback in each behavior record of the user; S1-2, find out the association rule of the second-order frequent itemsets according to the set minimum support and minimum confidence. 3. The recommendation method based on a hybrid algorithm of item collaborative filtering and association rules according to claim 1, wherein the step S2 comprises the following specific steps: S2-1. Construct a user-item scoring matrix according to the user's historical records and scoring records. Each row of the matrix is the evaluation score of different users on different items, and each column of the matrix is a different item; S2-2. Use the association rule in step 1 to predict the rating of an item that has had positive feedback with a certain user, and fill the rating into the above-mentioned user-item rating matrix to obtain a new user-item rating matrix; S2-3. Calculate the similarity between different items by using the cosine similarity formula according to the user-item rating matrix after filling the rating. 4. The recommendation method based on a hybrid algorithm of item collaborative filtering and association rules according to claim 1, wherein the step S3 comprises the following specific steps: S3-1. Construct all item-content attribute matrix, whether each row of the matrix has the content attribute of different items; S3-2. According to the item-content attribute matrix, the cosine similarity is used to calculate the similarity between different items. 5. The recommendation method based on a hybrid algorithm of item collaborative filtering and association rules according to claim 1, wherein the step S4 comprises the following specific steps: S4-1. The similarity between item i and item j is: sim(i,j) represents the similarity between item i and item j, the value range is [0,1], α represents the weight, the value range is [0.1, 0.9], N(i) represents the number of users who like item i , N(i)∩N(j) represents the number of users who like item i and item j at the same time, and V(i) represents the attribute vector of item i; S4-2, the calculation formula of item i prediction score is: P _ui represents user u’s predicted rating for item i, S(i, K) is the set of K items most similar to item i, and r _uj represents user u’s actual rating for item j; S4-3. Predict the rating of a user for an item according to a predictive rating formula.