Online shopping false marketing information identification method based on emotion divergence

Abstract

The invention belongs to the field of false information identification, and particularly relates to an online shopping false marketing information identification method based on emotional divergence, which comprises the following steps: acquiring user online shopping comment data and preprocessing to obtain preprocessed data; obtaining a subject emotion dependency factor of each comment text by adopting a unit emotion factor representation method; calculating an overall emotion dependence factor of each commodity according to the subject emotion dependence factor, and adjusting and updating the overall emotion dependence factor of each commodity by adopting a game theory; constructing a network relation graph according to the overall emotion dependency factor, and obtaining a user divergence degree through the network relation graph; obtaining a monthly average release amount and a monthly average coverage rate of user evaluation, and fusing the user divergence degree, the monthly average release amount and the monthly average coverage rate to obtain a user emotion diversity factor; and processing the user emotion diversity factor by adopting a sigmoid function to obtain an identification result.

Description

Technical Field

[0001] The present invention belongs to the field of false information identification, and in particular relates to a method for identifying false marketing information of online shopping based on sentiment divergence. Background Art

[0002] False marketing information mainly refers to untrue information that misleads consumers in commercial marketing activities. It is common in social life. In order to increase the sales of their products, merchants promote their products in the form of virtual buyers. In the information age, the proliferation of online false marketing information has become an inevitable phenomenon. Therefore, the identification of false information has received great attention and has become a hot topic in academic research. It can be seen that the research on the identification of online false marketing information is very important.

[0003] In recent years, many experts and scholars have done a lot of research in the field of false information identification. According to the different research focuses, these methods can be roughly divided into three categories: identification based on information content, identification based on user characteristics, and identification based on information dissemination characteristics.

[0004] During the research, we discovered an interesting phenomenon: false marketing information on online shopping platforms is often evaluation information issued by virtual buyers led by merchants. Through further research, we found that there is a significant difference in the degree of emotional divergence between the evaluation information posted by virtual buyers and actual buyers. For daily consumer goods in life, the degree of emotional divergence between the evaluation information posted by actual buyers and virtual buyers of merchants is more obvious. Combining the degree of emotional divergence in user comments and comprehensively considering the current status of false information identification research, we found that there are still the following challenges in the identification of false marketing information based on emotional divergence:

[0005] 1. The hidden nature of user evaluation emotions. For some user evaluation information, the types of emotions it contains are relatively rich, and there is no obvious sign of emotional tendency judgment. Therefore, the overall emotional tendency it contains cannot usually be accurately judged and classified. Therefore, how to accurately perceive and judge the emotions of the evaluation content is a very challenging task.

[0006] 2. Uncertainty in measuring the degree of sentiment divergence. The degree of sentiment divergence between different reviews posted by the same user may be affected by many factors, and the degree of sentiment divergence is not accurate enough in quantification. Therefore, it is difficult to design a model that takes into account the influence of various factors to accurately quantify the degree of sentiment divergence between different reviews.

[0007] 3. Computational complexity of evaluation information. Shopping information on shopping platforms is characterized by massive data, various types, and high complexity. Data processing often takes a lot of time. Therefore, it is still challenging to improve the efficiency of identifying false marketing information and make it more accurate. Summary of the invention

[0008] To solve the above problems, the present invention provides a method for identifying false marketing information for online shopping based on sentiment divergence, comprising the following steps:

[0009] S1. Obtain user online shopping review data and preprocess it to obtain preprocessed data; the preprocessed data includes a set of user reviews on different products, each product review set includes multiple review texts, and all the review texts are arranged in chronological order of release time;

[0010] S2. Based on the preprocessed data, the unit sentiment factor representation method is used to obtain the main sentiment dependency factor of each comment text;

[0011] S3. Calculate the overall emotional dependence factor of each commodity based on the subject emotional dependence factor, and adjust and update the overall emotional dependence factor of each commodity using game theory;

[0012] S4. Construct a network relationship diagram based on the overall sentiment dependency factor, and obtain the user divergence through the network relationship diagram;

[0013] S5. Obtain the average monthly release volume and average monthly coverage of user reviews, and integrate the user divergence, average monthly release volume, and average monthly coverage to obtain the user sentiment diversity factor;

[0014] S6. Use the sigmoid function to process the diverse factors of user emotions and obtain the recognition results.

[0015] Beneficial effects of the present invention:

[0016] Aiming at the problem that user emotions are hidden and cannot be accurately perceived, the present invention proposes a unit emotion factor characterization method to obtain the main emotion dependency factor of each comment text from the perspective of words. In this method, each evaluation text is divided into small units, and the unit emotion intensity of each small unit is obtained according to the word emotion intensity of each emotion word. The unit emotion dependency factor of each small unit is obtained by analyzing the mutual influence between small units, and the main emotion dependency factor of the evaluation unit body is calculated according to the unit emotion dependency factor, thereby better digging out the emotional tendency in the evaluation content and making more accurate judgment and classification.

[0017] Since the degree of emotional divergence between different reviews posted by users may be affected by many factors, and the degree of emotional divergence is not accurate enough in quantification, the present invention measures the degree of emotional divergence by constructing a network relationship graph, wherein the graph closeness centrality theory is introduced to better measure the degree of user divergence.

[0018] The present invention divides users into virtual buyers of merchants and actual buyers of customers, and realizes the key identification of false marketing information according to the degree of emotional divergence in shopping evaluations, thereby improving the accuracy of false marketing information identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a framework diagram of a false marketing information identification method based on user evaluation sentiment divergence according to the present invention;

[0020] Figure 2 This is a diagram showing the emotional game of user evaluation difference moment evaluation information of the present invention;

[0021] Figure 3 This is a quantitative diagram showing the degree of emotional divergence among users of the present invention regarding different commodities. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] The present invention provides a method for identifying false marketing information in online shopping based on sentiment divergence. Figure 1 As shown, the following steps are included:

[0024] S1. Obtain user online shopping review data and preprocess it to obtain preprocessed data; the preprocessed data includes a set of user reviews on different products, each product review set includes multiple review texts, and all the review texts are arranged in chronological order of release time.

[0025] S2. Based on the preprocessed data, the unit sentiment factor representation method is used to obtain the main sentiment dependency factor of each comment text.

[0026] Preferably, the present invention proposes a unit emotion factor characterization method, referred to as the UEFR method. The method uses text mining technology to split the evaluation text into multiple small units, then characterizes the emotional tendency of the small units through emotional dependency factors, and finally integrates the emotional dependency factors of each small unit to obtain the overall emotional tendency of the evaluation text, revealing the emotional tendency that may be implicit in the evaluation sentence.

[0027] In step S2, the subject sentiment dependency factor of each comment text is obtained through the UEFR method.

[0028] Specifically, a comment text is taken as an evaluation unit body, and step S2 processes an evaluation unit body, including:

[0029] S21. Use sentence boundary detection technology to divide the evaluation unit body into multiple small units;

[0030] S22. For each small unit, the word sentiment intensity of each sentiment word in the small unit is calculated through the BosonNLP sentiment dictionary, and the unit sentiment intensity of the small unit is obtained by adding up all the word sentiment intensities, which is expressed as

[0031]

[0032] Among them, Eins(w i ) represents the i-th sentiment word w in the small unit i The intensity of the associated emotion, Nunw(w i ) represents the i-th sentiment word w in the small unit i The number of associated negative words, SentiIns(w i ) represents the sentiment value of the i-th sentiment word itself; the present invention divides the intensity of the sentiment expressed by the adverb, QualDeg(w i ) represents the i-th sentiment word w in the small unit i the degree of the associated adverbs;

[0033] The unit sentiment intensity of a small unit is defined as

[0034]

[0035] Among them, Uein(v i ) represents a small unit v i Unit emotional intensity, Nuw(v i ) represents a small unit v i The number of sentiment words.

[0036] S22. Define the sentiment intensity of the conjunction between each two adjacent small units as follows

[0037]

[0038] Then the mutual influence between two adjacent small units is quantified as

[0039] Emi(vi,vj)=Emco(vi,vj)×log2(Uein(vi)×Uein(vj))

[0040] Among them, co(v i ,v j ) represents a small unit v i With small unit v j conjunction between i ,v j )∈Empr(c i ) represents a small unit v i With small unit v j The conjunctions between them belong to emotional progressive conjunctions, co(v i ,v j )∈Empa(c i ) represents a small unit v i With small unit v j The conjunction between the two is a conjunction of emotional transition, co(v i ,v j )∈Emtr(c i ) represents a small unit v i With small unit v j The conjunctions between are emotional substitution conjunctions; Emco (v i ,v j ) represents a small unit v i With small unit v j The emotional intensity of the conjunction between i ) represents a small unit v i The unit emotional intensity, Emi(v i ,v j ) represents a small unit v i With small unit v j The degree of mutual influence between

[0041] S23. The unit emotion dependency factor of each small unit is obtained according to the unit emotion intensity and mutual influence, as shown below:

[0042]

[0043] Among them, Nuc(v i ) indicates that the user is in the small unit v i The number of reviews published for the corresponding product, Nutl(v i ) represents the total number of user reviews.

[0044] The subject emotion dependency factor of the evaluation unit subject is calculated based on the unit emotion dependency factor.

[0045] Specifically, in order to express the emotional tendency of the evaluation unit body, the emotional tendency relationship between the evaluation unit body and the small unit is first expressed as follows:

[0046]

[0047] Among them, Esd(e n ) represents the nth evaluation text e published by the user n The subjective emotional dependence factor, Esd(v i ) represents the evaluation text e n Small and medium units i The unit emotion dependency factor, Nu(e n ) represents the evaluation text e n The number of small units.

[0048] S3. Calculate the overall emotional dependence factor of each commodity based on the subject emotional dependence factor, and use game theory to adjust and update the overall emotional dependence factor of each commodity.

[0049] Specifically, the overall sentiment dependence factor of each commodity is calculated based on the subject sentiment dependence factor, which is expressed as

[0050]

[0051] in, Indicates product m i The overall emotional dependence factor, Esd(e n ) represents the nth evaluation text e n The subject emotion dependence factor, N represents the product m i The number of corresponding evaluation texts, t1 represents the user's evaluation of product m i The time when the first review text is published, t n Indicates the user's preference for product m i The time when the nth review text is published, w represents the half-life time. By introducing the half-life function To simulate the change in the importance of users' evaluation of the same product over time.

[0052] Specifically, Figure 2 As shown, the process of adjusting and updating the overall emotional dependence factor of any commodity using game theory in step S3 includes:

[0053] S41. Define a game strategy including two situations: user evaluation support and user evaluation opposition. For each evaluation text corresponding to the product, if its subject emotion dependency factor is positive, the evaluation text is considered to be user evaluation support, otherwise it is considered to be user evaluation opposition;

[0054] S42. Define the profit function for the two cases of user evaluation support and user evaluation opposition, expressed as

[0055] Rev sup (e n)=Pa×Esd(e n ), Esd(e n )>0

[0056] Rev obj (e n )=Pb×Esd(e n ), Esd(e n )<0

[0057] Among them, Esd(e n ) represents the nth evaluation text e n The main emotional dependency factor of the product, Pa, Pb respectively represent the proportion of user evaluation support and user evaluation opposition in the user product evaluation, Pa + Pb = 1; Rev sup (e n ) indicates the revenue when the nth review text belongs to the user review support case, Rev obj (e n ) indicates the revenue when the nth review text belongs to the user's review objection;

[0058] S43. Calculate the sentiment influence of each comment text according to the profit function, where

[0059] If the comment text n If it belongs to the user evaluation support situation, its emotional influence is expressed as Mut sup (e n ), the calculation formula is

[0060]

[0061] If the comment text n If the user's evaluation is against the situation, its emotional influence is expressed as Mut obj (e n ), the calculation formula is

[0062]

[0063] S44. Update the overall emotional dependency factor according to the emotional influence, expressed as

[0064]

[0065] Among them, x, y are adjustment coefficients, N sup Indicates the number of evaluation texts that belong to user evaluation support, N obj Indicates the number of evaluation texts that are opposed by users, N sup +N obj =N, N represents product m i The number of corresponding evaluation texts; represents the adjusted overall emotional dependence factor, represents the overall affective dependency factor before adjustment.

[0066] We characterize users' emotional inclination towards a single product based on the overall emotional dependency factor adjusted by the product review set, providing a basis for quantifying users' emotional differences towards multiple products later.

[0067] S4. Construct a network relationship diagram based on the overall emotional dependency factors, and obtain user divergence through the network relationship diagram.

[0068] Preferably, the present invention proposes a commodity evaluation graph quantification model EGR. The model combines game theory to quantify the correlation between multiple evaluation emotions, and at the same time quantifies the degree of divergence between multiple commodity evaluation sentiment tendencies based on graph proximity centrality theory, such as Figure 3 shown.

[0069] Specifically, step S4 includes:

[0070] S41. In order to reflect the relationship between the overall emotional dependence factors of two different commodities, the difference between different commodities is defined as

[0071]

[0072] Among them, Dis(m i ,m j ) indicates product m i and product m j The difference between Indicates product m i The overall emotional dependence factor;

[0073] S42. Treat each product as a node, and assign the edge weight between each two nodes as the difference between the two products to obtain a network relationship graph. Based on this, we can quantify the degree of user sentiment divergence between different products by measuring the closeness of the network relationship graph.

[0074] S43. Calculate the closeness centrality of each node according to the closeness centrality theory of the graph, and calculate the average closeness centrality through the closeness centralities of all nodes;

[0075] Specifically, the calculation formula for the closeness centrality of each node is:

[0076]

[0077] Where Clc(i) represents the closeness centrality of node i; M represents the number of nodes, i.e. the number of commodities; Hvc(m i ) indicates product m iThe heat value objectively reflects the importance of the product in the minds of users. Based on this, we can get the overall average closeness centrality Clcto of all nodes in the graph:

[0078]

[0079] Through the above formula, we can judge the closeness of the network relationship graph as a whole, and then obtain the degree of divergence of users' emotional tendencies towards different products. Therefore, we use Clcto as an important indicator to quantify divergence.

[0080] S44. Calculate user divergence based on average closeness centrality, expressed as

[0081]

[0082] Through the above calculations, we can obtain the user's emotional divergence, which is an important basis for dividing users into merchant virtual buyers and user actual buyers.

[0083] S5. Obtain the average monthly release volume and average monthly coverage of user reviews, and integrate the user divergence, average monthly release volume, and average monthly coverage to obtain the user sentiment diversity factor.

[0084] S6. Use the sigmoid function to process the diverse factors of user emotions and obtain the recognition results.

[0085] Preferably, the present invention proposes a user evaluation feature association identification model BUAI. The model introduces the Yager combination rule and combines the user's evaluation frequency and the time distribution of the evaluation to accurately classify the user, thereby reducing the computational complexity of the evaluation information. In addition, the model also proposes a user emotion diversity factor to divide the user into a merchant virtual buyer and a customer actual buyer to distinguish false marketing information.

[0086] Get the average monthly release volume of user reviews Anp(u i At the same time, the monthly average coverage rate Mcr(u i ), the specific calculation is as follows:

[0087]

[0088] Dir(u i ) represents the average number of days per month involved in user reviews (i.e., the average number of days in a month that users posted reviews), and Dam represents the average number of days per month;

[0089] The above-mentioned multiple features are integrated together for comprehensive evaluation in combination with Yager combination rules, and the user emotion diversity factor is introduced at the same time. The specific calculation is as follows:

[0090] Usd(u i )=w1×Div(u i )+w2×Anp(u i )+w3×Mcr(u i )

[0091] a1, a2, a3 are credibility weights, and a1+a2+a3=1∩a1,a2,a3∈(0,1).

[0092] Combined with the sigmoid function, the probability value of the obtained combined feature model is predicted. The BUAI model can be expressed by the formula:

[0093] P = sigmoid[Usd(u i )]

[0094] in, is the activation function. We let the output of the model be expressed as P(n, fuser), which is defined as follows:

[0095]

[0096] Among them, θ represents the recognition result. If θ=1, it means that the current user is judged as a virtual buyer of the merchant, that is, the information he / she publishes is false marketing information; if θ=0, it means that the current user is judged as an actual buyer of the customer, that is, the information he / she publishes is real evaluation information; P(n,f|user) represents the output of the sigmoid function, max[P(n,f|user)] represents the maximum value of the output of the sigmoid function, and min[P(n,f|user)] represents the minimum value of the output of the sigmoid function.

[0097] In the present invention, unless otherwise clearly stipulated and limited, the terms such as "installation", "setting", "connection", "fixation" and "rotation" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral one; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium; it can be the internal connection of two elements or the interaction relationship between two elements. Unless otherwise clearly defined, ordinary technicians in this field can understand the specific meanings of the above terms in the present invention according to the specific circumstances.

[0098] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying false marketing information for online shopping based on sentiment divergence, characterized in that: The following steps are involved: S1. Obtain user online shopping review data and preprocess it to obtain preprocessed data; The preprocessed data includes user evaluation sets for different products. Each product evaluation set includes multiple review texts, and all review texts are arranged in the order of release time. S2. Based on the preprocessed data, the unit sentiment factor representation method is used to obtain the main sentiment dependency factor of each comment text; S3. Calculate the overall emotional dependence factor of each commodity based on the subject emotional dependence factor, and adjust and update the overall emotional dependence factor of each commodity using game theory; S4. Construct a network relationship diagram based on the overall sentiment dependency factor, and obtain the user divergence through the network relationship diagram; S5. Obtain the average monthly release volume and average monthly coverage of user reviews, and integrate the user divergence, average monthly release volume, and average monthly coverage to obtain the user sentiment diversity factor; S6. Use the sigmoid function to process the diverse factors of user emotions and obtain the recognition results.

2. According to claim 1, a method for identifying online shopping false marketing information based on emotional divergence is characterized in that: Taking a comment text as an evaluation unit body, step S2 processes an evaluation unit body, including: S21. Divide the evaluation unit into multiple small units; S22. For each small unit, calculate the word sentiment strength of each sentiment word in the small unit, and add up all the word sentiment strengths to obtain the unit sentiment strength of the small unit; S22. Calculate the mutual influence between every two adjacent small units, expressed as Emi(vi,vj)=Emco(vi,vj)×log2(Uein(vi)×Uein(vj)) Among them, co(v i ,v j ) represents a small unit v i With small unit v j conjunction between i ,v j )∈Empr(c i ) represents a small unit v i With small unit v j The conjunctions between them belong to emotional progressive conjunctions, co(v i ,v j )∈Empa(c i ) represents a small unit v i With small unit v j The conjunction between the two is a conjunction of emotional transition, co(v i ,v j )∈Emtr(c i ) represents a small unit v i With small unit v j The conjunctions between are emotional substitution conjunctions; Emco (v i ,v j ) represents a small unit v i With small unit v j The emotional intensity of the conjunction between i ) represents a small unit v i The unit emotional intensity, Emi(v i ,v j ) represents a small unit v i With small unit v j The degree of mutual influence between S23. The unit emotion dependency factor of each small unit is obtained according to the unit emotion intensity and the mutual influence, and the subject emotion dependency factor of the evaluation unit body is calculated according to the unit emotion dependency factor.

3. According to claim 1, a method for identifying online shopping false marketing information based on emotional divergence is characterized in that: The overall emotional dependence factor of each commodity is calculated based on the subject emotional dependence factor, expressed as in, Indicates product m i The overall emotional dependence factor, Esd(e n ) represents the nth evaluation text e n The subject emotion dependence factor, N represents the product m i The number of corresponding evaluation texts, t1 represents the user's evaluation of product m i The time when the first review text is published, t n Indicates the user's preference for product m i The moment when the nth review text is published, represents the half-life function, and w represents the half-life time.

4. According to claim 1, a method for identifying online shopping false marketing information based on emotional divergence is characterized in that: Step S3 uses game theory to adjust and update the overall emotional dependence factor of any commodity, including: S41. Define a game strategy including two situations: user evaluation support and user evaluation opposition. For each evaluation text corresponding to the product, if its subject emotion dependency factor is positive, the evaluation text is considered to be user evaluation support, otherwise it is considered to be user evaluation opposition; S42. Define the profit function for the two cases of user evaluation support and user evaluation opposition, expressed as Rev sup (and n )=Pa×Esd(e n ),Esd(e n )>0 Rev obj (and n )=Pb×Esd(e n ),Esd(e n )<0 Among them, Esd(e n ) represents the nth evaluation text e n The main emotional dependency factor of the user, Pa, and Pb respectively represent the proportion of user evaluation support and user evaluation opposition in user product evaluation. Pa+Pb=1;Rev sup (e n ) indicates the revenue when the nth review text is supported by user reviews. Rev obj (e n ) indicates the revenue when the nth review text belongs to the user's review objection; S43. Calculate the sentiment influence of each comment text according to the profit function, where If the comment text n If it belongs to the user evaluation support situation, its emotional influence is expressed as Mut sup (e n ), the calculation formula is If the comment text n If the user's evaluation is against the situation, its emotional influence is expressed as Mut obj (e n ), the calculation formula is S44. Update the overall emotional dependency factor according to the emotional influence, expressed as Among them, x, y are adjustment coefficients, N sup Indicates the number of evaluation texts that belong to user evaluation support, N obj Indicates the number of evaluation texts that are opposed by users, N sup +N obj =N, N represents product m i The number of corresponding evaluation texts; represents the adjusted overall emotional dependence factor, represents the overall affective dependency factor before adjustment.

5. The method for identifying online shopping false marketing information based on sentiment divergence according to claim 1, characterized in that: Step S4 specifically includes: S41. Calculate the difference between different products, expressed as Among them, Dis(m i ,m j ) indicates product m i and product m j The difference between Indicates product m i The overall emotional dependence factor; S42. Consider each commodity as a node, assign the edge weight between each two nodes to the difference between the two commodities, and obtain a network relationship diagram; S43. Calculate the closeness centrality of each node according to the closeness centrality theory of the graph, and calculate the average closeness centrality through the closeness centralities of all nodes; S44. Calculate user divergence based on average closeness centrality.

6. The method for identifying online shopping false marketing information based on sentiment divergence according to claim 5, characterized in that: The calculation formula of step S43 is: Where Clc(i) represents the closeness centrality of node i; M represents the number of nodes, that is, the number of commodities; Clcto represents the average closeness centrality, Hvc(m i ) indicates product m i The heat value.

7. The method for identifying online shopping false marketing information based on emotional divergence according to claim 1, characterized in that: The user emotion diversity factor is expressed as Usd(u i )=a1×Div(u i )+a2×Anp(u i )+a3×Mcr(u i ) Among them, Usd(u i ) represents the user's emotional diversity factor, Div(u i ) represents the user divergence, Anp(u i ) represents the average monthly number of user reviews, Mcr(u i ) represents the average monthly coverage of user reviews; Clcto represents the average closeness centrality, Dir(u i ) represents the average number of days per month involved in user evaluations, Dam represents the average number of days per month; a1, a2, a3 are credibility weights, and a1+a2+a3=1∩a1,a2,a3∈(0,1).

8. The method for identifying online shopping false marketing information based on sentiment divergence according to claim 1, characterized in that: In step S6, the recognition result is expressed as Among them, θ represents the recognition result. If θ=1, it means that the current user is judged as the merchant’s virtual buyer. If θ=0, it means that the current user is judged as the customer’s actual buyer. P(n,f|user) represents the output of the sigmoid function, max[P(n,f|user)] represents the maximum value of the sigmoid function output, and min[P(n,f|user)] represents the minimum value of the sigmoid function output.

Images