Collaborative filtering recommendation method based on attribute correlation enhancement

Abstract

The invention is applicable to the technical field of data recommendation, and provides a collaborative filtering recommendation method based on attribute correlation enhancement, which comprises the following steps: firstly, acquiring a user attribute table and a user-item score table; then, on the basis of the obtained user attribute table, defining and processing mutual relations between attributes, generating a mutual relation enhancement information table, defining a mutual association condition attribute similarity relation, and calculating a potential similar user set; calculating the similarity between the users by using an enhanced information table containing mutually associated condition attributes to obtain a similar user set; and finally, performing score prediction and recommendation on the similarity of the target users based on the similar users. According to the method, the null value * n is introduced to process the null value condition, and the similarity relation based on the correlation condition attributes is provided, so that a feasible method is provided for user similarity judgment and personalized recommendation based on the user attribute table under the condition that the attribute data is incomplete, and the recommendation effect under the user cold start scene is effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of data recommendation technology, and in particular relates to a collaborative filtering recommendation method based on enhanced attribute relationships. Background Technology

[0002] In the era of big data, recommender systems have become a key technology applied across various industries. With the rapid development of information technology, user behavior data generated in various fields is growing rapidly, making the effective processing and analysis of this massive amount of data a key research issue. Collaborative filtering, as an effective means of extracting potential information from large amounts of data, is being widely applied. Traditional collaborative filtering algorithms heavily rely on historical user behavior data to construct a set of similar users. However, when faced with new users without any rating records (cold start scenario), the lack of effective rating information makes it impossible to calculate the similarity between users, thus failing to obtain similar users for the target user.

[0003] Although similarity calculation methods based on user attributes can calculate user similarity using information such as gender, age, and region, user attribute tables often have problems such as missing attribute values ​​in practical applications. In this case, simply using user attribute tables with many default values ​​cannot obtain high-quality similar users, thus failing to obtain high-precision recommendation results. Summary of the Invention

[0004] The purpose of this invention is to provide a collaborative filtering recommendation method based on enhanced attribute relationships, aiming to solve the problems mentioned in the background art.

[0005] This invention is implemented as follows: a collaborative filtering recommendation method based on enhanced attribute relationships includes the following steps:

[0006] Step 1: Obtain the user attribute table and the user-item rating table;

[0007] Step 2: Based on the obtained User Attribute Table (UA), define and process the relationships between attributes, generate an enhanced relationship information table, define the similarity relationships of interrelated conditional attributes, and calculate the set of potential similar users;

[0008] Step 3: Using an enhanced information table containing interrelated conditional attributes, calculate the similarity between users to obtain a set of similar users;

[0009] Step 4: Based on the similarity between similar users and the target user, perform rating prediction and recommendation.

[0010] In a further technical solution, step 1 includes the following specific steps:

[0011] Step 1.1: Obtain the User Attribute Table UA, which represents multiple attribute information of the user. The attribute table is shown in Equation (1):

[0012] UA=(U,AT,{V a |a∈AT},B,ρ) (1)

[0013] Where U is a finite and non-empty set of users; AT is a finite and non-empty set of attributes; V a B is the range of any conditional attribute a ∈ A; B is defined in V a ×V b The binary relation; ρ is an information function, ρ:U×AT→V, that is, ρ(u,a)=V, which means that the value of user u under attribute a is V;

[0014] If the user attribute table UA contains at least one conditional attribute a∈AT with a value range V a If the value * is null, the User Attribute Table (UA) is considered incomplete.

[0015] Step 1.2: Obtain the User-Item Rating Table UR, which represents the user's rating information for each item. The UR is shown in Equation (2):

[0016] UR=(U,I,F) (2)

[0017] Where I is a finite and non-empty set of items; F is the user's rating of the items.

[0018] In a further technical solution, step 2 includes the following specific steps:

[0019] Step 2.1: Define and handle the relationships between attributes;

[0020] Suppose a, b∈AT are two conditional attributes with ranges V and V respectively. a and V b B is defined in V a ×V b If there exists a user u∈U such that (ρ(u,a),ρ(u,b))∈B holds, then it means that conditional attributes a and b are related to each other through the binary relation B; based on the binary relation B between conditional attributes a and b, the related conditional attributes are defined as aBb.

[0021] When the values ​​of conditional attributes a and b satisfy the binary relation B, the corresponding user set B(a,b) is represented as follows:

[0022] B(a,b)={u∈U|(ρ(u,a),ρ(u,b))∈B} (3)

[0023] When the relationship between a and b cannot be considered through B, the relationship between null value * and other values ​​is not handled through B.

[0024] If either a or b is empty, i.e., ρ(u,a) = * or ρ(u,b) = *, then the user set is represented as NULL(a,b):

[0025] NULL(a,b)={u∈U|ρ(u,a)=*}∪{u∈U|ρ(u,b)=*} (4)

[0026] At this point, the scope of B can be naturally extended to include the case where null values ​​* are not satisfied, including the user set B that does not satisfy the binary relation B. c (a,b) can be represented as:

[0027] B c (a,b)=U-(B(a,b)∪NULL(a,b)) (5)

[0028] Step 2.2: Generate a table of enhanced interrelationships;

[0029] From the user attribute information table UA and the interrelated conditional attributes aBb, generate an information table UA with enhanced interrelationships. int :

[0030] UA int =(U,AT) int V∪{0,1},B int ,ρ int (6)

[0031] Among them AT int =AT∪aBb; ρ int =U×AT int →V∪{0,1}∪NULL(a,b); ρ int The value of (u,c) is obtained from formula (7):

[0032]

[0033] Step 2.3: Define the similarity relationships of interrelated conditional attributes and calculate the set of potential similar users:

[0034] gather Represented as Based on the user attribute table, the potential similarity relationship between users is calculated using the potential similarity relationship formula, which is shown in equation (8) below:

[0035]

[0036] in It is a subset of attributes that includes interrelated conditional attributes, where u and v represent users on U;

[0037] Based on the potential similarity relation SIM(A), the set of potential similar users S for each user is directly calculated. A (u) = {v∈U|(u,v)∈SIM(A)}; that is, for each conditional attribute a∈A, check whether users u and v satisfy the similarity condition on this attribute, and add the user v that meets the condition to the set of potential similar users S of user u. A (u) in.

[0038] In a further technical solution, step 3 includes the following specific steps:

[0039] Using an information table enhanced with mutual relationships, based on the potential similar user set S A (u), calculate the similarity between user u and similar users v in the potential similar user set, and select the top k users with the highest similarity as the similar user set N(u) for calculating the predicted score of user u. The similarity calculation formula is shown in equation (9):

[0040]

[0041] Where sim(u,v) represents the similarity between user u and user v, |A| is the size of the attribute set, and match(u,v,a) is the degree of matching between users u and v on conditional attribute a, defined as:

[0042]

[0043] Such as the potential similar user set S A If the number of similar users in (u) is less than k, then the similarity of other users to user u is calculated, and the first k-|S are selected. A (u) | a user and a set of potential similar users S A (u) is the set of similar users N(u).

[0044] In a further technical solution, step 4 includes the following specific steps:

[0045] Step 4.1: Based on the ratings of item i by the similar user set N(u), predict the rating of item i by user u; the rating prediction formula is shown in formula (13):

[0046]

[0047] Where P(u,i) represents the predicted rating of user u for item i. Let SA(u) represent the average rating of all items from the set of similar users, sim(u,v) represent the similarity between user u and user v, and r(v,i) represent the actual rating of user v for item i. This represents the average rating of user v; if the calculated predicted rating is greater than 5, it is set to 5, and if it is less than 0, it is set to 0.

[0048] Step 4.2: Sort the predicted ratings of the unrated items and recommend the top n unrated items with the highest predicted ratings to the user.

[0049] This invention provides a collaborative filtering recommendation method based on enhanced attribute relationships, which introduces null values*. n This paper addresses the issue of null values ​​and proposes a similarity relationship based on interrelated conditional attributes. This provides a feasible method for determining user similarity and making personalized recommendations based on user attribute tables when attribute data is incomplete, effectively improving the recommendation effect in user cold start scenarios. Attached Figure Description

[0050] Figure 1 A flowchart illustrating a collaborative filtering recommendation method based on enhanced attribute relationships, provided in an embodiment of the present invention;

[0051] Figure 2 The recommendation accuracy (Precision) is adjusted as the number of neighboring users of the target user changes. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0053] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0054] like Figure 1 As shown, an embodiment of the present invention provides a collaborative filtering recommendation method based on enhanced attribute relationships, comprising the following steps:

[0055] Step 1: Obtain the user attribute table and the user-item rating table;

[0056] Step 1.1: Obtain the User Attribute Table UA, which represents multiple attribute information of the user. The attribute table is shown in Equation (1):

[0057] UA=(U,AT,{V a |a∈AT},B,ρ) (1)

[0058] Where U is a finite and non-empty set of users; AT is a finite and non-empty set of attributes; V a B is the range of any conditional attribute a ∈ A; B is defined in V a ×V b The binary relation; ρ is an information function, ρ:U×AT→V, that is, ρ(u,a)=V, which means that the value of user u under attribute a is V;

[0059] If the user attribute table UA contains at least one conditional attribute a∈AT with a value range V a If the value * is null, the User Attribute Table (UA) is considered incomplete.

[0060] Step 1.2: Obtain the User-Item Rating Table UR, which represents the user's rating information for each item. The UR is shown in Equation (2):

[0061] UR=(U,I,F) (2)

[0062] Where I is a finite and non-empty set of items; F is the user's rating of the items;

[0063] Step 2: Based on the obtained User Attribute Table (UA), define and process the relationships between attributes, generate an enhanced relationship information table, define the similarity relationships of interrelated conditional attributes, and calculate the set of potential similar users;

[0064] Step 2.1: Define and handle the relationships between attributes;

[0065] Suppose a, b∈AT are two conditional attributes with ranges V and V respectively. a and V b B is defined in V a ×V b A binary relation is defined as follows: if there exists a user u∈U such that (ρ(u,a),ρ(u,b))∈B holds, then conditional attributes a and b are related to each other through the binary relation B. Based on the binary relation B between conditional attributes a and b, the related conditional attributes are defined as aBb.

[0066] When the values ​​of conditional attributes a and b satisfy the binary relation B, the corresponding user set B(a,b) is represented as follows:

[0067] B(a,b)={u∈U|(ρ(u,a),ρ(u,b))∈B} (3)

[0068] If the relationship between a and b cannot be considered through B, then the relationship between null value * and other values ​​is not handled through B.

[0069] If either a or b is empty, i.e., ρ(u,a) = * or ρ(u,b) = *, then the user set is represented as NULL(a,b):

[0070] NULL(a,b)={u∈U|ρ(u,a)=*}∪{u∈U|ρ(u,b)=*} (4)

[0071] At this point, the scope of B can be naturally extended to include the case where null values ​​* are not satisfied, including the user set B that does not satisfy the binary relation B. c (a,b) can be represented as:

[0072] B c (a,b)=U-(B(a,b)∪NULL(a,b)) (5)

[0073] Step 2.2: Generate a table of enhanced interrelationships;

[0074] From the user attribute information table UA and the interrelated conditional attributes aBb, generate an information table UA with enhanced interrelationships. int :

[0075] UA int =(U,AT) int V∪{0,1},B int ,ρ int (6)

[0076] Among them AT int =AT∪aBb; ρ int =U×AT int →V∪{0,1}∪NULL(a,b); ρ int The value of (u,c) is obtained from formula (7):

[0077]

[0078] Step 2.3: Define the similarity relationships of interrelated conditional attributes and calculate the set of potential similar users:

[0079] gather Represented as Based on the user attribute table, the potential similarity relationship between users is calculated using the potential similarity relationship formula, which is shown in equation (8) below:

[0080]

[0081] in It is a subset of attributes that includes interrelated conditional attributes, where u and v represent users on U.

[0082] Based on the potential similarity relation SIM(A), the set of potential similar users S for each user is directly calculated. A (u) = {v∈U|(u,v)∈SIM(A)}. That is, for each conditional attribute a∈A, check whether users u and v satisfy the similarity condition on that attribute (they have the same value on conditional attribute a, or one of the users has a null value on conditional attribute a). Add the user v that meets the condition to the set S of potential similar users of user u. A (u) in.

[0083] Step 3: Using an enhanced information table containing interrelated conditional attributes, calculate the similarity between users to obtain a set of similar users;

[0084] Using an information table enhanced with mutual relationships, based on the potential similar user set S A (u), calculate the similarity between user u and similar users v in the potential similar user set, and select the top k users with the highest similarity as the similar user set N(u) for calculating the predicted score of user u. The similarity calculation formula is shown in equation (9):

[0085]

[0086] Where sim(u,v) represents the similarity between user u and user v, |A| is the size of the attribute set, and match(u,v,a) is the degree of matching between users u and v on conditional attribute a, defined as:

[0087]

[0088] Such as the potential similar user set S A If the number of similar users in (u) is less than k, then the similarity of other users to user u is calculated, and the first k-|S are selected. A (u) | a user and a set of potential similar users S A (u) is the set of similar users N(u).

[0089] Step 4: Based on the similarity between similar users and the target user, predict and recommend ratings;

[0090] Step 4.1: Based on the ratings of item i by similar user set N(u), predict the rating of item i by user u. The rating prediction formula is shown in formula (13):

[0091]

[0092] Where P(u,i) represents the predicted rating of user u for item i. Let SA(u) represent the average rating of all items from the set of similar users, sim(u,v) represent the similarity between user u and user v, and r(v,i) represent the actual rating of user v for item i. This represents the average rating for user v. A predicted rating greater than 5 is assigned the value 5, and less than 0 is assigned the value 0.

[0093] Step 4.2: Sort the predicted ratings of the unrated items and recommend the top n unrated items with the highest predicted ratings to the user.

[0094] As a preferred embodiment of the present invention, taking a user-rated movie rating information table as an example, the method is performed according to the following steps:

[0095] Step 1: Obtain the user attribute table and user item rating table;

[0096] Step 1.1: Obtain the user attribute table Table 1. Table 1 is a user attribute table containing 8 users (u1-u8) and 4 conditional attributes (gender, education level, credit score, and income). Here, U represents the user set; Gender represents the user's gender; Education represents the user's education level; CreditScore and Income represent the user's credit rating and income level, respectively; * indicates a null value; and vg indicates "very good". The presence of null values ​​in this table is consistent with the typical situation of user attribute tables and the main application scenario of this method.

[0097] Table 1

[0098] U Gender Education Credit Score Income u1 female bachelor good v.g. u2 male master good v.g. u3 male master * good u4 female bachelor normal normal u5 female * * * u6 male master good * u7 male master * bad u8 * bachelor good normal

[0099] Step 1.2: Obtain user project rating table 2, which contains user rating information for the product.

[0100] Table 2

[0101]

[0102]

[0103] The "-" indicates that the user has not yet rated the product.

[0104] Step 2: Based on the obtained user attribute table, define and process the relationships between attributes, generate an enhanced relationship information table, define the similarity relationships of interrelated conditional attributes, and calculate the potential similar user set;

[0105] Step 2.1 Define and process the relationships between attributes;

[0106] Taking the relationship between credit rating and income level as an example, we introduce the binary relation "better than" (<), indicating that the income level rating is better than the credit rating. Here, <(CreditScore,Income) is the support set of the relationship between credit rating and income level based on the binary relation "better than" (<), that is, the set of objects that support the relation "CreditScore<Income". Based on the binary relation "better than" (<), we use "CreditScore≤Income" to represent the relationship where a user's income level is "better than or equal to" their credit rating.

[0107] Based on set B(a,b), B c The definitions of (a,b) and NULL(a,b) are used to calculate:

[0108] <(CreditScore,Income)={u1,u2}

[0109] < c (CreditScore,Income)={u4,u8}

[0110] NULL(CreditScore,Income)={u3,u5,u6,u7}

[0111] Step 2.2: Generate a table of enhanced interrelationships;

[0112] According to formula (7), the "CreditScore < Income" value for u1 and u2 is 1; the "CreditScore < Income" value for u4 and u8 is 0; and the "CreditScore < Income" value for u3, u5, u6, and u7 is *. n The results are summarized in Table 3:

[0113] Table 3

[0114]

[0115]

[0116] Where CS stands for CreditScore and I stands for Income.

[0117] Step 2.3: Define the similarity relationships of interrelated conditional attributes and calculate the set of potential similar users:

[0118] Taking the relationship between credit score and income as “better than” (<) as an example, using the latent similarity formula (8), the following latent similarity relationship SIM (CS<I) is obtained:

[0119] For users u1 and u2: u1, u2 ∈ < (CS, I), that is, both support that the income score is better than the credit score, so (u1, u2) ∈ SIM (CS < I), indicating that u1 and u2 have potential similarity in this relationship;

[0120] For users u4 and u8: u4, u8 ∈ < c(CS, I), that is, neither of them supports the income score being better than the credit score. Therefore, (u4, u8) ∈ SIM(CS < I), indicating that u4 and u8 have potential similarity in this relationship.

[0121] For users u3 and u7: u3, u7 ∈ NULL (CS, I), both belong to the case where the credit score is null but the income is not null. Therefore, (u3, u7) ∈ SIM (CS < I), indicating that u3 and u7 have potential similarity in this relationship.

[0122] Based on the enhanced information table of mutual relationships, select the attribute set A = {Gender, CS < I, CS ≤ I}, and calculate the set of potential similar users S for each user. A (u):

[0123] S A (u1) = {u5}

[0124] S A (u2) = {u3, u6, u7}

[0125] S A (u3) = {u2, u6, u7}

[0126] S A (u4)={u5}

[0127] S A (u5) = {u1, u4}

[0128] S A (u6) = {u2, u3, u7}

[0129] S A (u7) = {u2, u3, u6}

[0130]

[0131] Step 3: Calculate the similarity between users using an enhanced information table containing interrelated conditional attributes.

[0132] Suppose user u5 is the target user, and the set of similar users S of u5 is... A In (u5), the similarity between all similar users and u5 is calculated according to the similarity formula (9), and the top k=3 users with the highest similarity are selected. Excluding u5 itself, S AIf there are fewer than 3 similar users to u5 in (u5), an additional set of similar users S is calculated. A The similarity between all users other than u5 and u5 is calculated, and the user with the highest similarity is added to the similar user set N(u5).

[0133] sim(u5,u1)=(1.0+0.5+0.5) / 3=0.667

[0134] sim(u5,u2)=(0+0.5+0.5) / 3≈0.333

[0135] sim(u5,u3)=(0+0.3+0.3) / 3=0.200

[0136] sim(u5,u4)=(1.0+0.5+0.5) / 3=0.667

[0137] sim(u5,u6)=(0+0.3+0.3) / 3=0.200

[0138] sim(u5,u7)=(0+0.3+0.3) / 3=0.200

[0139] sim(u5,u8)=(0.5+0.5+0.5) / 3=0.500

[0140] Therefore, users u1, u4, and u8 are selected as similar users of u5, and N(u5) = {u1, u4, u8}.

[0141] Step 4: Based on the similarity between similar users and the target user, predict and recommend ratings;

[0142] Step 4.1: Recommend financial products that user u5 may be interested in. Based on the ratings of product i by the similar user set N(u5) of user u5, use formula (13) to obtain the predicted rating formula for product i by user u5:

[0143]

[0144] Based on the user project rating table in Table 2, the product rating table for user u5 and similar users of u5 is shown in Table 4:

[0145] Table 4

[0146] U Product A Product B Product C Product D Product E u1 5 3 4 4 5 u4 3 3 1 5 4 u5 - - - - - u8 - 4 5 - 4

[0147] The "-" indicates that the user has not yet rated the product.

[0148] For target user u5, the predicted score is calculated using the similar user set N(u5)={u1,u4,u8} and the previously calculated corresponding similarity.

[0149] For each product that user u5 has not rated, substitute the product into the predicted rating formula to calculate the predicted rating:

[0150] P(u5,A)=(5+3) / 2+[(0.667×(5-4.2)+0.667×(3-3.2))] / 1.834≈4.22

[0151] P(u5,B)=(3+3+4) / 3+[(0.667×(3-4.2)+0.667×(3-3.2)+0.500×(4-4.33))] / 1.834≈2.73

[0152] P(u5,C)=(4+1+5) / 3+[(0.667×(4-4.2)+0.667×(1-3.2)+0.500×(5-4.33))] / 1.834≈2.64

[0153] P(u5,D)=(4+5) / 2+[(0.667×(4-4.2)+0.667×(5-3.2))] / 1.834≈5

[0154] P(u5,E)=(5+4+4) / 3+[(0.667×(5-4.2)+0.667×(4-3.2)+0.500×(4-4.33))] / 1.834≈4.82

[0155] Step 4.2: Sort the unrated items according to the predicted ratings of all unrated items for user u5, and recommend the top 3 financial products with the highest predicted ratings to user u5;

[0156] Based on the predicted scores, the predicted scores are: Product D (5) > Product E (4.82) > Product A (4.22) > Product B (2.73) > Product C (2.64).

[0157] Therefore, we recommend three products to user u5: product D, product E, and product A.

[0158] As a preferred embodiment of the present invention, experimental verification of this method is conducted, specifically including:

[0159] (1) Using public datasets

[0160] The dataset used in the validation process was MovieLens. This dataset contains 943 users, 1682 movies, and 100,000 ratings, with rating values ​​distributed as {1, 2, 3, 4, 5}. Each user rated at least 20 movies. The user attribute table includes user ID, gender, age, occupation, and other attribute information.

[0161] (2) Evaluation indicators

[0162] Precision is used to measure the accuracy of recommendations. The higher the value, the higher the accuracy of the recommendations. The formula for calculating precision is as follows:

[0163]

[0164] In the formula, M represents the number of users, and TP i This represents the number of items that were actually interacted with by the i-th user and were subsequently recommended. N represents the number of items in the recommendation list.

[0165] (3) Comparison and analysis of experimental results

[0166] The experimental results were compared with those of the traditional user-based collaborative filtering (UBCF) algorithm, which directly utilizes user attributes in the dataset when calculating user similarity. Figure 2 The comparison results are shown. The results demonstrate that the precision of this method is significantly higher than that of UBCF. Since a higher precision value corresponds to higher recommendation accuracy, this method enhances the accuracy of similar user calculations in cold start scenarios by mining the interrelationships between user attributes, effectively improving recommendation precision.

[0167] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A collaborative filtering recommendation method based on enhanced attribute relationships, characterized in that, Includes the following steps: Step 1: Obtain the User Attribute Table (UA) and the User-Item Rating Table; Step 2: Based on the obtained User Attribute Table (UA), define and process the relationships between attributes, generate an enhanced relationship information table, define the similarity relationships of interrelated conditional attributes, and calculate the set of potential similar users; Step 3: Using an enhanced information table containing interrelated conditional attributes, calculate the similarity between users to obtain a set of similar users; Step 4: Based on the similarity between similar users and the target user, perform rating prediction and recommendation.

2. The collaborative filtering recommendation method based on enhanced attribute relationships according to claim 1, characterized in that, Step 1 includes the following specific steps: Step 1.1: Obtain the User Attribute Table UA, which represents multiple attribute information of the user. The attribute table is shown in Equation (1): UA=(U,AT,{V a |a∈AT},B,ρ) (1) Where U is a finite and non-empty set of users; AT is a finite and non-empty set of attributes; V a B is the range of any conditional attribute a ∈ A; B is defined in V a ×V b The binary relation; ρ is an information function, ρ:U×AT→V, that is, ρ(u,a)=V, which means that the value of user u under attribute a is V; If the user attribute table UA contains at least one conditional attribute a∈AT with a value range V a If the value * is null, the User Attribute Table (UA) is considered incomplete. Step 1.2: Obtain the User-Item Rating Table UR, which represents the user's rating information for each item. The UR is shown in Equation (2): UR=(U,I,F) (2) Where I is a finite and non-empty set of items; F is the user's rating of the items.

3. The collaborative filtering recommendation method based on enhanced attribute relationships according to claim 2, characterized in that, Step 2 includes the following specific steps: Step 2.1: Define and handle the relationships between attributes; Suppose a, b∈AT are two conditional attributes with ranges V and V respectively. a and V b B is defined in V a ×V b If there exists a user u∈U such that (ρ(u,a),ρ(u,b))∈B holds, then it means that conditional attributes a and b are related to each other through the binary relation B; based on the binary relation B between conditional attributes a and b, the related conditional attributes are defined as aBb. When the values ​​of conditional attributes a and b satisfy the binary relation B, the corresponding user set B(a,b) is represented as follows: B(a,b)={u∈U|(ρ(u,a),ρ(u,b))∈B} (3) When the relationship between a and b cannot be considered through B, the relationship between null value * and other values ​​is not handled through B. If either a or b is empty, i.e., ρ(u,a) = * or ρ(u,b) = *, then the user set is represented as NULL(a,b): NULL(a,b)={u∈U|ρ(u,a)=*}∪{u∈U|ρ(u,b)=*} (4) At this point, the scope of B can be naturally extended to include the case where null values ​​* are not satisfied, including the user set B that does not satisfy the binary relation B. c (a,b) can be represented as: B c (a,b)=U-(B(a,b)∪NULL(a,b)) (5) Step 2.2: Generate a table of enhanced interrelationships; From the user attribute information table UA and the interrelated conditional attributes aBb, generate an information table UA with enhanced interrelationships. int : UA int =(U,AT int ,V∪{0,1},B int ,ρ int ) (6) Among them AT int =AT∪aBb; ρ int =U×AT int →V∪{0,1}∪NULL(a,b); ρ int The value of (u,c) is obtained from formula (7): Step 2.3: Define the similarity relationships of interrelated conditional attributes and calculate the set of potential similar users: gather Represented as Based on the user attribute table, the potential similarity relationship between users is calculated using the potential similarity relationship formula, which is shown in equation (8) below: in It is a subset of attributes that includes interrelated conditional attributes, where u and v represent users on U; Based on the potential similarity relation SIM(A), the set of potential similar users S for each user is directly calculated. A (u) = {v∈U|(u,v)∈SIM(A)}; that is, for each conditional attribute a∈A, check whether users u and v satisfy the similarity condition on this attribute, and add the user v that meets the condition to the set of potential similar users S of user u. A (u) in.

4. The collaborative filtering recommendation method based on enhanced attribute relationships according to claim 2, characterized in that, Step 3 includes the following specific steps: Using an information table enhanced with mutual relationships, based on the potential similar user set S A (u), calculate the similarity between user u and similar users v in the potential similar user set, and select the top k users with the highest similarity as the similar user set N(u) for calculating the predicted score of user u. The similarity calculation formula is shown in equation (9): Where sim(u,v) represents the similarity between user u and user v, |A| is the size of the attribute set, and match(u,v,a) is the degree of matching between users u and v on conditional attribute a, defined as: Such as the potential similar user set S A If the number of similar users in (u) is less than k, then the similarity of other users to user u is calculated, and the first k-|S are selected. A (u) | a user and a set of potential similar users S A (u) is the set of similar users N(u).

5. The collaborative filtering recommendation method based on enhanced attribute relationships according to claim 4, characterized in that, Step 4 includes the following specific steps: Step 4.1: Based on the ratings of item i by the similar user set N(u), predict the rating of item i by user u; the rating prediction formula is shown in formula (13): Where P(u,i) represents the predicted rating of user u for item i. Let SA(u) represent the average rating of all items from the set of similar users, sim(u,v) represent the similarity between user u and user v, and r(v,i) represent the actual rating of user v for item i. This represents the average rating of user v; if the calculated predicted rating is greater than 5, it is set to 5, and if it is less than 0, it is set to 0. Step 4.2: Sort the predicted ratings of unrated items and recommend the top n unrated items with the highest predicted ratings to the user.

Images