User similarity calculation method, calculation system, device and storage medium
By combining global and local similarity to calculate user similarity, the problems of data sparsity and cold start in the user-product rating matrix are solved, resulting in more accurate user similarity assessment and more precise product recommendations.
Patent Information
- Application Number
- CN202111123866.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-09-24
- Publication Date
- 2026-01-16
- Estimated Expiration
- 2041-09-24
AI Technical Summary
When existing user similarity calculations are based on user-item rating matrices, data sparsity and cold start issues lead to inaccurate calculations of user similarity, affecting the reliability and accuracy of the recommendation system.
User similarity is calculated by combining global and local similarity. The weighted sum of global and local similarity is used to calculate local similarity through an improved Bach coefficient and a modified cosine similarity, making full use of all rating information.
When there is limited shared rating information among users, this method improves the accuracy of user similarity calculation, alleviates the limitations of data sparsity and cold start, and enhances the precision of product recommendations.
Smart Images
Figure CN114004275B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present disclosure belongs to the technical field of electronic commerce, and particularly relates to a user similarity calculation method, a user similarity calculation system, a computer device, and a computer readable storage medium. BACKGROUND
[0002] A good recommendation system can promote consumer consumption behavior, improve user shopping experience and increase retention, and ultimately achieve the purpose of increasing the transaction amount of the seller. In the recommendation system, the collaborative filtering (CF) algorithm is a representative algorithm and is widely used by various e-commerce platforms. The collaborative filtering algorithm mainly includes a user-based collaborative filtering (User-CF) algorithm and an item-based collaborative filtering (Item-CF) algorithm. As shown in the following formula (1), the key of the User-CF algorithm is to find similar users of a target user, and to recommend preferred goods of the similar users to the target user. The steps are divided into three steps: 1, obtaining user-goods rating information; 2, calculating user similarity according to the user-goods rating information and sorting according to the size, and taking the top N users with higher similarity as a neighbor user set; and 3, rating prediction for unknown goods of the user according to the neighbor user set, and recommending the goods with higher predicted rating to the user. Figure 1
[0003] It can be seen that the user similarity calculation is the key of the User-CF algorithm. The user similarity calculation is completed based on a user-goods rating matrix, and the strategies that can be used for solving include cosine similarity, modified cosine similarity, Pearson correlation coefficient, and Jaccard similarity.
[0004] Since the existing user similarity calculation is completed based on the user-goods rating matrix, it needs to be centralized, and there is enough user behavior information. When the user historical behavior is less, there is not enough common goods rating information between users, that is, the user-goods rating matrix data is sparse. For the rating of non-commonly rated goods, the existing user similarity calculation cannot be used, which leads to inaccurate similarity calculation between users, and thus it is difficult to make recommendations with high accuracy. The recommendation reliability and recommendation accuracy of the recommendation system are not high. SUMMARY
[0005] The present disclosure provides a user similarity calculation method, a calculation system, a computer device, and a storage medium, which can calculate the user similarity even when there is less common rating information between users, alleviate the limitations of data sparsity and cold start, and make the goods recommendation more accurate.
[0006] In a first aspect, the present disclosure provides a method for calculating user similarity, comprising:
[0007] obtaining a user-item rating matrix;
[0008] calculating global similarity between any two users in the user-item rating matrix by using a preset global similarity calculation method, and calculating local similarity between any two users in the user-item rating matrix by using a preset local similarity calculation method;
[0009] calculating overall similarity between any two users in the user-item rating matrix according to the global similarity and the local similarity between the two users.
[0010] Further, the global similarity between any two users is calculated by using the following formula:
[0011]
[0012] In formula (1), sim global (u,v) is the global similarity between user u and user v in the user-item rating matrix, R ui represents the rating of user u to item i, R vj represents the rating of user v to item j, respectively represents the average rating of all rated items by user u and user v, I u , I v respectively represents the rated items by user u and user v, I uv represents the common rated items by user u and user v, i and j are positive integers.
[0013] Further, the local similarity between any two users is calculated based on the modified Bhattacharyya coefficient, wherein the modified Bhattacharyya coefficient is calculated by using the following calculation formula:
[0014]
[0015] In formula (2), BC'(u,v) is the modified Bhattacharyya coefficient between user u and user v in the user-item rating matrix, which represents the user represents the total number of rated items by user u, represents the total number of rated items by user v, represents the total number of items rated as k by user u or user v, and m represents the total number of item rating categories.
[0016] Further, when there is no common rated item between any two users in the user-item rating matrix, the local similarity of the any two users is calculated by the following formula:
[0017] sim local (u, v) = BC'(u, v) (3)
[0018] In formula (3), sim local (u, v) is the local similarity of user u and user v in the user-item rating matrix, and BC'(u, v) is the modified Bhattacharyya coefficient of user u and user v in the user-item rating matrix.
[0019] Further, when there is a common rated item between any two users in the user-item rating matrix, the local similarity of the any two users is calculated by the following formula:
[0020]
[0021] In formula (4), sim local (u, v) is the local similarity of user u and user v in the user-item rating matrix, BC'(u, v) is the modified Bhattacharyya coefficient of user u and user v in the user-item rating matrix, R ui represents the rating of item i by user u, R vj represents the rating of item j by user v, respectively represent the average rating of all rated items by user u and user v, I u , I v respectively represent the rated items of user u and user v, I uv represent the common rated items of user u and user v, and i and j are positive integers.
[0022] Further, the overall similarity of any two users in the user-item rating matrix is calculated by the global similarity and the local similarity of the any two users by the following formula:
[0023] sim(u, v) = sim global (u, v) + sim local (u, v) (5)
[0024] In formula (5), sim(u, v) is the overall similarity of user u and user v in the user-item rating matrix, sim global (u, v) is the global similarity of user u and user v, and sim local (u, v) is the local similarity of user u and user v.
[0025] In a second aspect, the present disclosure provides a user similarity calculation system, comprising:
[0026] an acquisition module configured to acquire a user-item rating matrix;
[0027] a first calculation module configured to calculate global similarity between any two users in the user-item rating matrix using a preset global similarity calculation method;
[0028] a second calculation module configured to calculate local similarity between any two users in the user-item rating matrix using a preset local similarity calculation method;
[0029] a third calculation module configured to calculate overall similarity between any two users in the user-item rating matrix according to the global similarity and the local similarity between the two users.
[0030] Further, the third calculation module is specifically configured to:
[0031] calculate the overall similarity between any two users in the user-item rating matrix using the following formula (5):
[0032] sim(u, v) = sim global (u, v) + sim local (u, v) (5)
[0033] In formula (5), sim(u, v) is the overall similarity between user u and user v in the user-item rating matrix, sim global (u, v) is the global similarity between user u and user v, and sim local (u, v) is the local similarity between user u and user v.
[0034] In a third aspect, the present disclosure further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the user similarity calculation method according to any one of the first aspect.
[0035] In a fourth aspect, the present disclosure further provides a computer readable storage medium comprising a computer program, which, when running on a computer, causes the computer to execute the user similarity calculation method according to any one of the first aspect.
[0036] Advantages:
[0037] The user similarity calculation method, the user similarity calculation system, the computer device and the storage medium provided by the present disclosure are characterized in that: a user-goods score matrix is obtained; global similarity of any two users in the user-goods score matrix is calculated by using a preset global similarity calculation method, and local similarity of the any two users is calculated by using a preset local similarity calculation method; and overall similarity of the any two users is calculated according to the global similarity and the local similarity of the any two users in the user-goods score matrix. The technical solution of the present disclosure fully utilizes all the score information, comprehensively considers the global similarity and the local similarity, so that the user similarity can be calculated even if the common score information between users is less, and the limitations of data sparsity and cold start are alleviated. In the local similarity calculation, an improved Bhattacharyya coefficient is introduced, the score distribution rule of the user is comprehensively considered, and the problem of different numbers of goods scored by two users is solved, so that the similarity calculation result is more in line with the actual situation, and the goods recommendation is more accurate. BRIEF DESCRIPTION OF DRAWINGS
[0038] Figure 1 A schematic diagram of a user-based collaborative filtering recommendation algorithm in the prior art;
[0039] Figure 2 A flowchart of a user similarity calculation method provided by the first embodiment of the present disclosure;
[0040] Figure 3 An architecture diagram of a user similarity calculation system provided by the second embodiment of the present disclosure;
[0041] Figure 4 An architecture diagram of a computer device provided by the third embodiment of the present disclosure. DETAILED DESCRIPTION
[0042] In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the present disclosure will be further described in detail below with reference to the drawings and embodiments.
[0043] The terms used in the embodiments of the present disclosure are merely for the purpose of describing specific embodiments, and are not intended to limit the present disclosure. The singular forms "a", "an" and "the" used in the embodiments of the present disclosure and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.
[0044] Since the existing user similarity calculation is based on the user-item rating matrix, it needs sufficient user behavior information in the data set. When the user's historical behavior is less, there is not enough common item rating information between users, that is, the user-item rating matrix data is sparse. For the rating of non-common rating items, the existing user similarity calculation cannot be used, which leads to inaccurate similarity calculation between users, and it is difficult to make accurate recommendations with high accuracy, and the recommendation reliability and recommendation accuracy are not high.
[0045] The technical solutions of the present disclosure and how the technical solutions of the present disclosure solve the above problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.
[0046] Figure 2 A flowchart of a user similarity calculation method in a collaborative filtering algorithm provided by Embodiment One of the present disclosure is shown in Figure 2 , which includes:
[0047] Step S101: Obtain a user-item rating matrix;
[0048] Step S102: According to the user-item rating matrix, calculate the global similarity of any two users by using a pre-set global similarity calculation method, and
[0049] Step S103: Calculate the local similarity of any two users by using a pre-set local similarity calculation method;
[0050] Step S104: Calculate the overall similarity of any two users according to the global similarity and the local similarity of the two users in the user-item rating matrix.
[0051] User similarity calculation is the key to User-CF algorithm. User similarity calculation is based on user-item rating matrix, for example, there is a user-item rating matrix R mn As follows:
[0052]
[0053] In the formula, m represents m users, n represents n items, R mn represents the rating of the mth user to the nth item, and user similarity calculation is performed using row vectors. The strategies that can be used in solving include cosine similarity, modified cosine similarity, Pearson correlation coefficient, Jaccard similarity, Bhattacharyya coefficient, etc.
[0054] Specifically, (1) cosine similarity calculation (set the rating value of the item without rating to 0):
[0055]
[0056] In the formula, R ui R vi Let I represent the ratings of user u and user v for product i, respectively. u I v Let I represent the sets of items rated by users u and v, respectively. uv This refers to the products that users u and v rated together.
[0057] (2) Modified cosine similarity calculation (set the rating value to 0 for products without ratings):
[0058]
[0059] In the formula, Rui represents the rating of user u for product i, and R vj This represents user v's rating of product j. Let I represent the average ratings of users u and v for the rated items, respectively. u I v I represents the products rated by users u and v, respectively. uv This refers to the products that users u and v rated together.
[0060] (3) Calculation of Pearson correlation coefficient:
[0061]
[0062] In the formula, R ui R represents the rating of user u for product i. vj This represents user v's rating of product j. Let I represent the average ratings of users u and v for the rated items, respectively. uv This refers to the products that users u and v rated together.
[0063] The modified cosine similarity is calculated similarly to the Pearson correlation coefficient, but the difference is that the Pearson correlation coefficient is based on the similarity of products with shared user ratings, while the modified cosine similarity is based on the similarity of all products.
[0064] (4) Jaccard similarity:
[0065] Jaccard similarity does not consider the user's rating of a product, but only whether the user has a preference for the product, that is, the proportion of product ratings shared by two users to their total ratings. The value is between (0, 1), where a value of 0 indicates that the two users have no common preference, and a value of 1 indicates that the two users have the same preference.
[0066]
[0067] where I u , I v denote the item sets rated by user u and user v, respectively.
[0068] 2. Jaccard Coefficient
[0069] Jaccard Coefficient is a measure of the similarity between two rating vectors. On continuous distributions, the Jaccard Coefficient is defined as On discrete distributions, the Jaccard Coefficient is defined as
[0070] Since the user-item rating is discrete, the similarity between user u and user v can be measured by the following formula:
[0071]
[0072] where m denotes the total number of categories of user-item ratings, where, denotes the total number of items rated by user u with rating k, denotes the total number of items rated by user u, and so on.
[0073] For users, if only the common ratings are used to calculate the similarity between users, this will lead to inaccurate similarity calculation between users in the case of less common ratings, and it is difficult to make accurate recommendations with high accuracy. The embodiments of the present disclosure calculate the overall similarity between users by adding local similarity, and no longer rely on the common ratings of users. Various factors that may affect the similarity between users are considered, all rating information is fully utilized, and the similarity between two users can be calculated even without common ratings, thereby alleviating the limitations of sparse rating matrix data and improving the accuracy of similarity calculation between users. The global similarity can be calculated by using the existing similarity calculation method, and the local similarity can be calculated by improving the existing similarity. The embodiments of the present disclosure calculate the local similarity by using the improved Jaccard Coefficient and the modified cosine similarity. The premise of calculating the local similarity by using the modified cosine similarity is that the rating value of the item without rating in the user-item rating matrix is set to 0. The Jaccard Coefficient compensates for the defect that the modified cosine similarity is not sensitive to the difference between numerical values. Then, the overall similarity between any two users in the user-item rating matrix is calculated according to the global similarity and the local similarity. All ratings of users on items are fully utilized, and the real similarity between users can be better reflected.
[0074] Further, the preset global similarity calculation method is used to calculate the global similarity of any two users, and the following formula is used to obtain the global similarity:
[0075]
[0076] In formula (1), Sim global (u, v) is the global similarity of user u and user v in the user-item rating matrix, R ui represents the rating of user u to item i, R vj represents the rating of user v to item j, respectively represent the average rating of all rated items of user u and user v, I u , I v respectively represent the rated items of user u and user v, I uv represent the common rated items of user u and user v, i and j are positive integers.
[0077] In formula (1), is a Pearson correlation coefficient calculation formula, is a Jaccard similarity calculation formula, the Jaccard similarity calculates the proportion of common ratings, and does not consider the absolute rating. The Pearson correlation coefficient needs to be substituted into the specific rating when calculating. The combination of the two can complement each other, so that the calculated global similarity is more consistent with the interest similarity between actual users.
[0078] Further, the local similarity of any two users is calculated based on the modified Bhattacharyya coefficient, wherein the modified Bhattacharyya coefficient is calculated by using the following calculation formula:
[0079]
[0080] In formula (2), BC'(u, v) is the modified Bhattacharyya coefficient of user u and user v in the user-item rating matrix, which represents the user represents the total amount of rated items of user u, represents the total amount of rated items of user v, represents the total amount of items rated as k by user u or user v, and m represents the total number of item rating categories.
[0081] Further, when any two users in the user-item rating matrix have no common rated items, the local similarity of the any two users is calculated by using the following calculation formula:
[0082] sim local (u, v) = BC'(u, v) (3)
[0083] sim(u, v) = BC'(u, v) (3) local (u, v) is the local similarity between user u and user v in the user-item rating matrix, BC'(u, v) is the modified Bhattacharyya coefficient between user u and user v in the user-item rating matrix.
[0084] When there is no common rated item between users, the global similarity cannot be calculated, and the local similarity uses the modified Bhattacharyya coefficient; the traditional Bhattacharyya coefficient formula does not consider the case where the number of rated items of two users is different, for example, user rating all scores are 1, 2, 3, 4, 5, user u's ratings for items A and B are 1 and 3 respectively, and user v's ratings for items A, B, C, and D are 3, 1, 3, and 1 respectively, user u's rating vector is (1, 3), and user v's rating vector is (3, 1, 3, 1), according to the traditional Bhattacharyya coefficient calculation formula,
[0085]
[0086] is the result of complete similarity, which is obviously inconsistent with the actual situation.
[0087] The improved calculation result is
[0088]
[0089] It can be seen that the result is obviously improved and more consistent with the actual situation. The actual calculation process only considers the probability distribution of the same rating number, and does not consider whether it is a common rating, aiming to evaluate the similarity of the rating law.
[0090] Further, when there is a common rated item between any two users in the user-item rating matrix, the local similarity between the two users is obtained by using the following calculation formula:
[0091]
[0092] sim(u, v) = BC'(u, v) (3) local (u, v) is the local similarity between user u and user v in the user-item rating matrix, BC'(u, v) is the modified Bhattacharyya coefficient between user u and user v in the user-item rating matrix, R ui represents the rating of user u for item i, R vj represents the rating of user v for item j, respectively represent the average rating of user u and user v for all rated items, I u , I v respectively represent the rated items of user u and user v, I uv represent the common rated items of user u and user v, i and j are positive integers.
[0093] In formula (4), sim The modified cosine similarity is calculated by setting the score value to 0 for the goods without scores in the user-goods score matrix, and the beta coefficient compensates for the defect that the modified cosine similarity is not sensitive to the difference between numerical values. When calculating, all scored goods of the user u and the user v are considered, not only the goods scored by both of them.
[0094] Further, the overall similarity between any two users in the user-goods score matrix is calculated according to the global similarity and the local similarity between the two users, and the following formula is used to obtain the overall similarity:
[0095] sim (u, v) = sim global (u, v) + sim local (u, v) (5)
[0096] In formula (5), sim (u, v) is the overall similarity between the user u and the user v in the user-goods score matrix, sim global (u, v) is the global similarity between the user u and the user v, and sim local (u, v) is the local similarity between the user u and the user v.
[0097] When the common scoring goods information of the users u and v is 0, the similarity between the users can still be calculated, and the result is equal to the local similarity; when the users u and v have common scoring goods information, the similarity is divided into two parts, the global similarity and the local similarity.
[0098] The embodiments of the present disclosure make full use of all the scoring information, comprehensively consider the global similarity and the local similarity, so that the user similarity can be calculated even when the common scoring information between users is less, and the limitations of data sparsity and cold start are alleviated. Among them, the improved beta coefficient is introduced in the calculation of the local similarity, which comprehensively considers the scoring distribution rule of the user and solves the problem of different number of goods scored by two users, so that the similarity calculation result is more in line with the actual situation, and the goods recommendation is more accurate.
[0099] Figure 3 The architecture diagram of a user similarity calculation system provided in the second embodiment of the present disclosure is shown in FIG. 1, which includes: Figure 3
[0100] The acquisition module 11 is configured to acquire a user-goods score matrix;
[0101] The first calculation module 12 is configured to calculate the global similarity between any two users in the user-goods score matrix according to a preset global similarity calculation method.
[0102] a second computing module 13 configured to calculate the local similarity between any two users in the user-item rating matrix according to a preset local similarity calculation method;
[0103] a third computing module 14 configured to calculate the overall similarity between any two users in the user-item rating matrix according to the global similarity and the local similarity between the two users.
[0104] Further, the first computing module 12 is specifically configured to calculate the global similarity between any two users according to the following formula:
[0105]
[0106] In formula (1), sim global (u, v) is the global similarity between user u and user v in the user-item rating matrix, R ui is the rating of item i by user u, R vj is the rating of item j by user v, respectively, is the average rating of all rated items by user u and user v, I u , I v respectively, is the rated item by user u and user v, I uv is the common rated item by user u and user v, i and j are positive integers.
[0107] Further, the second computing module 13 is specifically configured to calculate the local similarity between any two users based on the modified Bhattacharyya coefficient, wherein the modified Bhattacharyya coefficient is calculated according to the following formula:
[0108]
[0109] In formula (2), BC'(u, v) is the modified Bhattacharyya coefficient between user u and user v in the user-item rating matrix, indicating the user is the total number of rated items by user u, is the total number of rated items by user v, is the total number of items rated as k by user u or user v, and m is the total number of item rating categories.
[0110] Further, the second computing module 13 is specifically configured to calculate the local similarity between any two users in the user-item rating matrix according to the following formula when there is no common rated item between the two users:
[0111] sim local(u, v) = BC'(u, v) (3)
[0112] In formula (3) : sim local (u, v) is the local similarity of user u and user v in the user-item rating matrix, and BC'(u, v) is the modified B coefficient of user u and user v in the user-item rating matrix.
[0113] Further, the second calculation module 13 is specifically further configured to calculate the local similarity of any two users in the user-item rating matrix by using the following calculation formula when there is a common rated item between the any two users in the user-item rating matrix:
[0114]
[0115] In formula (4) : sim local (u, v) is the local similarity of user u and user v in the user-item rating matrix, BC'(u, v) is the modified B coefficient of user u and user v in the user-item rating matrix, R ui represents the rating of user u to item i, R vj represents the rating of user v to item j, respectively represents the average rating of user u and user v to all rated items, I u , I v respectively represents the rated item of user u and user v, I uv represents the common rated item of user u and user v, i and j are positive integers.
[0116] Further, the third calculation module 14 is specifically configured to:
[0117] The overall similarity of any two users in the user-item rating matrix is calculated by using the following formula (5) :
[0118] sim(u, v) = sim global (u, v) + sim local (u, v) (5)
[0119] In formula (5) : sim(u, v) is the overall similarity of user u and user v in the user-item rating matrix, sim global (u, v) is the global similarity of user u and user v, sim local (u, v) is the local similarity of user u and user v.
[0120] The user similarity calculation system of the embodiments of the present disclosure is used to implement the user similarity calculation method in method embodiment one, and thus the description is relatively simple. For details, refer to the related description in method embodiment one above, which will not be repeated here.
[0121] In addition, as Figure 4 shown, the third embodiment of the present disclosure also provides a computer device, including a memory 10 and a processor 20, the memory 10 stores a computer program, when the processor 20 runs the computer program stored in the memory 10, the processor 20 executes the above-mentioned various possible user similarity calculation methods.
[0122] In addition, the embodiments of the present disclosure also provide a computer readable storage medium, the computer readable storage medium stores computer execution instructions, when at least one processor of the user equipment executes the computer execution instructions, the user equipment executes the above-mentioned various possible methods.
[0123] The computer readable medium includes a computer storage medium and a communication medium, and the communication medium includes any medium that facilitates the transfer of computer program from one place to another. The storage medium can be any available medium that can be accessed by a general or special purpose computer. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC. In addition, the ASIC can be located in the user equipment. Of course, the processor and the storage medium can also exist as discrete components in the communication device.
[0124] It can be understood that the above embodiments are only exemplary embodiments adopted for illustrating the principles of the present disclosure, but the present disclosure is not limited thereto. Various modifications and improvements can be made by those skilled in the art without departing from the spirit and essence of the present disclosure, and these modifications and improvements are also considered to be within the protection scope of the present disclosure.
Claims
1. A method of calculating user similarity, characterized by, The method comprises the following steps: obtaining a user-goods rating matrix; calculating global similarity between any two users in the user-goods rating matrix by using a preset global similarity calculation method, and calculating local similarity between any two users in the user-goods rating matrix by using a preset local similarity calculation method; calculating overall similarity between any two users in the user-goods rating matrix according to the global similarity and the local similarity between the two users; the preset local similarity calculation method comprises: when there is no common rated goods between any two users in the user-goods rating matrix, directly using a modified Pielou coefficient as the local similarity; and when there is common rated goods, using a product of the modified Pielou coefficient and a modified cosine similarity as the local similarity; wherein the modified Pielou coefficient is calculated by using the following calculation formula: In formula (2): BC'(u, v) is the modified Bhattacharyya coefficient of user u and user v in the user-item rating matrix, indicating the similarity between user u and user v indicates the total number of items rated by user u, indicates the total number of items rated by user v, indicates the total number of items rated by user u or user v as k, and m indicates the total number of categories of items rated by the user.
2. The computational method of claim 1, wherein, the global similarity between any two users in the user-goods rating matrix is calculated by using the following formula: In formula (1), sim global (u, v) is the global similarity between user u and user v in the user-item rating matrix, R ui represents the rating of user u to item i, R vj represents the rating of user v to item j, respectively represents the average rating of user u, user v to all rated items, I u , I v respectively represents the rated items of user u, user v, I uv represents the common rated items of user u, user v, i, j are positive integers.
3. The computational method of claim 1, wherein, the product of the modified Pielou coefficient and the modified cosine similarity as the local similarity comprises: the local similarity between any two users is calculated by using the following calculation formula: In formula (4): sim local (u, v) is the local similarity of user u and user v in the user-item rating matrix, BC'(u, v) is the modified Bhattacharyya coefficient of user u and user v in the user-item rating matrix, R ui represents the rating of user u to item i, R vj represents the rating of user v to item j, respectively represent the average rating of user u and user v to all rated items, I u , I v respectively represent the rated items of user u and user v, I uv represent the common rated items of user u and user v, i, j are positive integers.
4. The calculation method according to any one of claims 1 to 3, characterized in that, the overall similarity between any two users in the user-goods rating matrix is calculated according to the global similarity and the local similarity between the two users by using the following formula: sim(u, v) = sim global (u, v) + sim local (u, v) (5) In formula (5): sim(u, v) is the overall similarity of user u and user v in the user-item rating matrix, sim global (u, v) is the global similarity of user u and user v, sim local (u, v) is the local similarity of user u and user v.
5. A user similarity calculation system, characterized by, The method comprises the following steps: an obtaining module is configured to obtain a user-goods rating matrix; a first calculating module is configured to calculate global similarity between any two users in the user-goods rating matrix by using a preset global similarity calculation method according to the user-goods rating matrix; a second calculating module is configured to calculate local similarity between any two users in the user-goods rating matrix by using a preset local similarity calculation method according to the user-goods rating matrix; a third calculating module is configured to calculate overall similarity between any two users in the user-goods rating matrix according to the global similarity and the local similarity between the two users; the preset local similarity calculation method comprises: when there is no common rated goods between any two users in the user-goods rating matrix, directly using a modified Pielou coefficient as the local similarity; and when there is common rated goods, using a product of the modified Pielou coefficient and a modified cosine similarity as the local similarity; wherein the modified Pielou coefficient is calculated by using the following calculation formula: In formula (2): BC'(u, v) is the modified Bhattacharyya coefficient of user u and user v in the user-item rating matrix, indicating the similarity between user u and user v indicates the total number of items rated by user u, indicates the total number of items rated by user v, indicates the total number of items rated by user u or user v as k, and m indicates the total number of categories of items rated by the user.
6. The computing system of claim 5, wherein, the third calculating module is specifically configured to: calculate the overall similarity between any two users in the user-goods rating matrix by using the following formula (5): sim(u, v) = sim global (u, v) + sim local (u, v) (5) In formula (5): sim(u, v) is the overall similarity of user u and user v in the user-item rating matrix, sim global (u, v) is the global similarity of user u and user v, sim local (u, v) is the local similarity of user u and user v.
7. A computer device, comprising: The computer program, when executed on a computer, causes the computer to perform the user similarity calculation method according to any one of claims 1-4.
8. A computer-readable storage medium comprising: The computer program, when executed on a computer, causes the computer to perform the user similarity calculation method according to any one of claims 1-4.