Personalized recommendation method based on language value two-tuple concept lattice

Abstract

The invention belongs to a data mining and intelligent information processing technology, and particularly relates to a personalized recommendation method based on a language value two-tuple concept lattice. The personalized recommendation method can solve the problems of fuzzy recommendation interpretation and cold start and can process different data types. The method comprises the following steps: collecting and preprocessing data, collecting data information in a multi-type data form, and judging whether the types of fuzzy data information collected by a computer are unified or not. If not, converting the data type into a language value two-tuple; constructing a language value two-tuple concept of the user set and the article set; obtaining language value two-tuple evaluation data of the user on the commodity decision attribute set of the commodity set; constructing a cognitive system of the training data set; constructing a sufficient knowledge base and a fuzzy object language knowledge simulation lattice of the training data set; constructing a necessary knowledge base and a fuzzy object language knowledge simulation lattice of the training data set; and extracting rules, and performing selection judgment to obtain a recommendation result.

Description

Technical Field

[0001] The present invention belongs to data mining and intelligent information processing technology, and in particular to a personalized recommendation method based on a language value binary concept lattice that can solve the recommendation interpretation fuzziness and cold start problems and can process different data types. Background Art

[0002] As an important tool for data analysis and information processing, formal concept analysis has been widely used in machine learning, data mining, information retrieval and other fields. Due to its ability to analyze the relationship between objects and attributes and generate concepts in a formal context, concept lattices have great advantages in rule extraction, and multiple rules can be obtained through certain algorithms. Using language values ​​to express the intrinsic properties between concepts can effectively reduce the information loss caused by fuzzy and uncertain data, making concept lattices better used in uncertain knowledge processing and data mining. Introducing language value models, especially language value binary models with data offsets, into concept lattices to obtain rules with language value information has become an important field of artificial intelligence, especially an issue worthy of in-depth study in the direction of intelligent information processing.

[0003] In real life, due to the complexity and uncertainty of human thinking, people's descriptions of things often exist in an imprecise and incomplete form. Professor Zadeh first proposed the fuzzy set theory in 1965, and used this form to deal with uncertain information in real life. In response to complex qualitative problems that are difficult to evaluate with precise information, Zadeh used the concept of language variables to model language information and produced fuzzy language methods. The analysis, processing and application of language value information have aroused the research interest of scholars. Xu proposed a variety of generalized induced language aggregation operators and studied the relationship between each aggregation operator. Chen et al. proposed a proportional hesitant fuzzy language term set and studied its basic aggregation operator as a special method for language distribution evaluation. The proportional information of each generalized language term is used to express its possibility distribution under different assumptions.

[0004] Since fuzzy language methods will cause certain information loss in the process of analysis and approximation, the final result of language information after fusion and discretization may deviate from the original cognition, resulting in errors and lack of accuracy. Therefore, based on the fuzzy language calculation method, Herrera et al. proposed a language value binary representation model, using symbol offset to supplement language term information. This model allows language value information to be expressed continuously in its domain, improving accuracy. Based on the hierarchical analysis method and fuzzy binary, Santos et al. proposed a decision support system that can integrate the qualitative analysis and quantitative performance of experts. Zhao et al. proposed a new TODIM method based on the cumulative prospect theory and binary language value set, which fully considers the psychology of decision makers and ensures relatively objective attribute weights.

[0005] At present, rule extraction has attracted many scholars to conduct research and has been widely used in many fields such as machine learning, medical diagnosis, and financial investment. Because the relationship between objects and attributes can be clearly reflected, concept lattices have great advantages in rule extraction and have become an effective tool for association rule extraction. Godin et al. first studied the use of concept lattices to extract association rules and proposed a classic algorithm for extracting corresponding rules. Shao et al. defined discernible matrices and discernible functions, and constructed a knowledge lossless method to reduce complexity in the context of decision forms, ensuring that the maximum rules extracted from the simplified decision form context are the same as those extracted from the initial decision form context.

[0006] However, the existing association rule extraction algorithm based on concept lattice applied to the recommendation system still has the problems of fuzzy recommendation interpretation and cold start. In addition, since the concept lattice still cannot handle different fuzzy data, it is easy to cause information processing difficulties. Summary of the invention

[0007] The present invention belongs to data mining and intelligent information processing technology, and in particular to a personalized recommendation method based on a language value binary concept lattice that can solve the recommendation interpretation fuzziness and cold start problems and can process different data types.

[0008] The technical solution of the present invention is: a personalized recommendation method based on a language value binary concept lattice, which is performed according to the following steps:

[0009] A. Data collection and preprocessing:

[0010] A1. Collect data information in multiple data forms and determine whether the fuzzy data information types collected by the computer are unified. If not, the data types include fuzzy sets. Intuitionistic Fuzzy Sets and hesitant fuzzy sets Suppose the language term set is S = {s0,s1...,s g}, α is a symbol offset, and the commodity attribute set is L = {l 1 ,l 2 ,…,l n}, the product set is U = {u1,u2,…,u m}, n is the total number of commodity attributes, and m is the total number of commodities;

[0011] A2. Collect user feedback on product u r Use a data type to describe the item i The vague concept u r ∈U,l i ∈L, initialize the formal background of the product set U and the item attribute set L As a training set is the fuzzy binary relationship from the commodity set U to the commodity attribute set L, that is,

[0012] A3. The data types in are converted to language value tuples according to the following conversion steps:

[0013] Judging data samples type;

[0014] If a[j] is a fuzzy set, then

[0015] W:[0,1]→S×[-0.5,0.5)

[0016]

[0017] If a[j] is an intuitionistic fuzzy set, then

[0018] G:[0,1]×[0,1]→S×[-0.5,0.5)

[0019]

[0020] If a[j] is an intuitionistic fuzzy set, then

[0021] P:H(U)→S×[-0.5,0.5)

[0022]

[0023] Formal background based on product set U and product attribute set L All binary relations in the context are converted into linguistic value binary pairs.

[0024] B. Define operations on the product set U and the product attribute set L: If X ⊙ =B⊙ At the same time B ⊙ =X ⊙ Then (X,B) is a two-tuple language concept.

[0025] Construct the language value tuple concept of user set U and item set L;

[0026] C. Obtain the user's language value binary evaluation data on the product set U and the product decision attribute set D (if it is not a language value binary evaluation, convert it according to step A3) to form a language value binary form background

[0027] C1. Cognitive stage: based on training data set and form Construct training data set using X operator and B fuzzy operator cognitive system;

[0028] C2. Based on training data set use definition Constructing training data set The sufficient knowledge base K1 and the language knowledge quasi-form composed of K1;

[0029] C3. Based on training data set use definition Constructing training data set The necessary knowledge base K2 and the language knowledge quasi-lattice composed of K2;

[0030] D. Personalized recommendation stage:

[0031] D1. Based on the training data set Using (X,B) is The concept of (Y, C) is The concept of X=Y, where Y≠U,φ, we get the rule B→C

[0032] D2. Analyze the extracted rules, make selection judgments, and obtain recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a structural diagram of the fuzzy object language knowledge pseudo-lattice constructed in part K1 of the embodiment of the present invention.

[0034] Figure 2 This is a structural diagram of the fuzzy object language knowledge pseudo-lattice constructed in part K2 of the embodiment of the present invention.

[0035] Figure 3 This is a flow chart of a recommendation algorithm for an embodiment of the present invention. DETAILED DESCRIPTION

[0036] Taking the selection of Spring Festival commodities as an example, the personalized recommendation method based on the language value binary concept lattice of the present invention is as follows: Figure 3 As shown, follow the steps below:

[0037] A. Data collection and preprocessing:

[0038] A1. Let the language term set be S = {s0,s1...,s g}, α is a symbol offset, and the commodity attribute set is L = {l 1 ,l 2 ,…,l n}, the product set is U = {u1,u2,…,u m}, n is the total number of commodity attributes, and m is the total number of commodities;

[0039] Before the Chinese New Year, people are used to buying decorations, hoping for good luck and prosperity in the new year. In order to participate in the Spring Festival promotion, users need to buy a product in a decoration store in a shopping system. Currently, there are 6 kinds of decorations participating in the event in this store, namely small colored lights, lanterns, New Year paintings, ornaments, fortune trees and Chinese knots. We have collected evaluation information on the material, style, color and firmness of these 6 kinds of decorations in the store, as well as information on users' attention to and willingness to buy new products. The types of evaluation data collected by users on the four features of material, style, color and firmness are not all in the form of language values, but also include other types of multi-type fuzzy data. All the fuzzy evaluation information of decorations is shown in Table 1.

[0040] Table 1 Fuzzy evaluation information of decoration features

[0041]

[0042] Let the language term set S = {s0, s1..., s6} represent the different evaluation levels of users on the products. The granularity of this language term set is 7, where s0 to s6 represent "very bad", "very bad", "bad", "average", "good", "very good", and "very good" respectively.

[0043] The collected information about user consumption tendency, that is, the evaluation information of users’ attention to new products and purchase intention, is in the form of language value binary tuples. The specific information is shown in Table 2.

[0044] Table 2 Consumption tendency evaluation information

[0045]

[0046] A2. Collect user feedback on product u rUse a data type to describe the item i The vague concept u r ∈U,l i ∈L, initialize the formal background of the product set U and the item attribute set L As a training set is the fuzzy binary relationship from the commodity set U to the commodity attribute set L, that is,

[0047] A3. The data types in are converted to language value tuples according to the following conversion steps:

[0048] Judging data samples type;

[0049] If a[j] is a fuzzy set, then

[0050] W:[0,1]→S×[-0.5,0.5)

[0051]

[0052] If a[j] is an intuitionistic fuzzy set, then

[0053] G:[0,1]×[0,1]→S×[-0.5,0.5)

[0054]

[0055] If a[j] is an intuitionistic fuzzy set, then

[0056] P:H(U)→S×[-0.5,0.5)

[0057]

[0058] Formal background based on product set U and product attribute set L All binary relations in are converted into language value binary form background, and the results are shown in Table 3.

[0059] Table 3 Decoration feature evaluation language value binary form background

[0060]

[0061] B. Define operations on the product set U and the product attribute set L: If X ⊙ =B ⊙ At the same time B ⊙ =X ⊙Then (X, B) is a two-tuple language concept, and the language value two-tuple concept of user set U and item set L is constructed; 1#(Φ,{a(s5,-0.4),b(s5,0.4),c(s4,0.1),d(s4,-0.4)})

[0062] 2#({6},{a(s5,-0.4),b(s5,0.4),c(s6,0.1),d(s4,0.0)})

[0063] 3#({4,6},{a(s5,-0.4),c(s4,0.4),d(s4,-0.4)})

[0064] 4#({5,6},{a(s5,-0.4),b(s5,0.4),c(s5,0.2)})

[0065] 5#({1,3,6},{a(s5,-0.4),b(s5,0.4),d(s4,-0.4)})

[0066] 6#({1,5,6},{a(s5,-0.4),b(s5,0.4)})

[0067] 7#({4,5,6},{a(s5,-0.4),c(s4,0.4)})

[0068] 8#({1,3,4,6},{a(s5,-0.4),d(s4,0.0)})

[0069] 9#({1,3,5,6},{a(s5,-0.4),b(s5,0.4)})

[0070] 10#({2,4,5,6},{c(s4,0.1)})

[0071] 11#({1,3,4,5,6},{a(s5,-0.4)})

[0072] 12#({1,2,3,4,5,6},Φ)

[0073] C. Obtain the user's language value binary evaluation data on the product set U and the product decision attribute set D (if it is not a language value binary evaluation, convert it according to step A3) to form a language value binary form background

[0074] C1. Cognitive stage: based on training data set and form Construct training data set using X operator and B fuzzy operator The cognitive system, the obtained language value binary decision form background is shown in Table 4;

[0075] Table 4 Decoration evaluation information language value binary decision form background

[0076]

[0077] C2. Based on training data set use definition Constructing training data set The sufficient knowledge base K1 and the language knowledge quasi-form composed of K1;

[0078] C3. Based on training data set use definition Constructing training data set The necessary knowledge base K2 and the language knowledge quasi-lattice composed of K2;

[0079] Since there is too much fuzzy object language knowledge in the knowledge base K1 and the knowledge base K2, the embodiment of the present invention only provides part of the sufficient knowledge base K1 and part of the necessary knowledge base K2. Figure 1 and Figure 2 As shown;

[0080] D. Personalized recommendation stage:

[0081] D1. Based on the training data set Using (X,B) is The concept of (Y, C) is The concept satisfies X=Y, where Y≠U,φ, and we get the rule B→C.

[0082] D2. Analyze the extracted rules, make selection judgments, and obtain recommendation results.

[0083] From the rule {abcd}→{ef}, we can get the following information: if the user requires the Chinese knots to be of rich styles, beautiful colors, good materials and firmness, and the material is slightly higher than the firmness, then he will be very willing to buy the Chinese knots of the store and pay more attention to the subsequent new products. In addition, due to e(s5,-0.2)

Claims

1. A personalized recommendation method based on a concept lattice of linguistic value binary tuples, characterized in that Follow these steps: A. Data collection and preprocessing: A1. Collect data information in multiple data forms and determine whether the fuzzy data information types collected by the computer are unified; if not, determine whether the data types include fuzzy sets. Intuitionistic Fuzzy Sets and hesitant fuzzy sets Suppose the language term set is S = {s0,s1...,s g }, α is a symbol offset, and the commodity attribute set is L = {l 1 ,l 2 ,…,l n }, the product set is U = {u1,u2,…,u m }, n is the total number of commodity attributes, and m is the total number of commodities; A2. Collect user feedback on product u r Use a data type to describe the item i The vague concept u r ∈U,l i ∈L, initialize the formal background of the product set U and the item attribute set L As a training set is the fuzzy binary relationship from the commodity set U to the commodity attribute set L, that is, A3. The data types in are converted to language value tuples according to the following conversion steps: Judging data samples type; If a[j] is a fuzzy set, then W:[0,1]→S×[-0.5,0.5) If a[j] is an intuitionistic fuzzy set, then G:[0,1]×[0,1]→S×[-0.5,0.5) If a[j] is an intuitionistic fuzzy set, then P:H(U)→S×[-0.5,0.5) Formal background based on product set U and product attribute set L All binary relations in the context are converted into linguistic value binary pairs; B. Define operations on the product set U and the product attribute set L: If X ⊙ =B ⊙ At the same time B ⊙ =X ⊙ Then (X,B) is a two-tuple language concept, Construct the language value tuple concept of user set U and item set L; C. Obtain the user's language value binary evaluation data on the product set U and the product decision attribute set D (if it is not a language value binary evaluation, convert it according to step A3) to form a language value binary form background C1. Cognitive stage: based on training data set and form Construct training data set using X operator and B fuzzy operator cognitive system; C2. Based on training data set use definition Constructing training data set The sufficient knowledge base K1 and the language knowledge quasi-form composed of K1; C3. Based on training data set use definition Constructing training data set The sufficient knowledge base K2 and the language knowledge quasi-form composed of K2; D. Personalized recommendation stage: D1. Based on the training data set Using (X,B) is The concept of (Y, C) is The concept satisfies X=Y, where Y≠U,φ, and we get the rule B→C; D2. Analyze the extracted rules, make selection judgments, and obtain recommendation results.

Images