An e-commerce recommendation method based on false review detection optimization

Abstract

This invention relates to the field of e-commerce recommendation systems, and particularly to an e-commerce recommendation method optimized based on fake review detection. The method includes acquiring user behavior data, user review data, and user-product interaction data; constructing a sentiment-behavior vectorization model, a sentiment-behavior matching detection model, and a sequence graph neural network recommendation model based on user credibility; and obtaining a recommendation list weakened by fake reviews based on the user behavior data, user review data, user-product interaction data, and each model. This invention quantifies the matching degree between user sentiment and behavior and uses it as an indicator of user credibility, increasing the influence of highly credible user behavior on the recommendation results, thereby effectively weakening the influence of fake review data and improving the accuracy of the e-commerce platform recommendation system.

Description

Technical Field

[0001] This invention relates to the field of e-commerce recommendation systems, specifically to an e-commerce recommendation method optimized based on fake review detection. Background Technology

[0002] With the rapid development of e-commerce, the recommendation system of online shopping platforms has become one of the important bases for users' purchasing decisions.

[0003] Currently, e-commerce platform recommendation systems mainly rely on user behavior data and review data for prediction. Traditional recommendation systems rely on data such as users' historical behavior and ad click-through rates for recommendations. However, fake reviews often manipulate recommendation algorithms through inauthentic emotional expressions or behavioral patterns, causing the system to be unable to effectively identify and filter this data, which in turn leads to a decrease in the accuracy of the system's recommendation results.

[0004] Therefore, reducing the impact of mismatched emotions and behaviors in fake reviews on recommendation systems has become one of the major challenges for e-commerce recommendation systems. Summary of the Invention

[0005] In view of this, the present invention proposes an e-commerce recommendation method based on fake review detection optimization to solve the above problems. The data processing of the e-commerce recommendation method based on fake review detection optimization described in this invention includes:

[0006] S1. Obtain user behavior data, user review data, and user-product interaction data from e-commerce platforms;

[0007] S2. Construct a sentiment-behavior vectorization model; input user behavior data and user comment data into the sentiment-behavior vectorization model to obtain behavior vectors and sentiment vectors;

[0008] S21. Obtain the set of historical behavior sequence features of users from user behavior data, B;

[0009] S22. Embed B to obtain the behavior vector B representing user behavior. emb ;

[0010] S23. Construct and optimize the emotion coding module;

[0011] S231. Replace user comment data with word sequences, and process the word sequences to select candidate words;

[0012] S232. Cluster the candidate words using the DBSCAN clustering method to obtain the sentiment seed word set;

[0013] S233. Use a sentiment seed word set to identify sentiment words in user comment data, and further identify matching attribute words based on sentiment words. Attribute words and sentiment words together form attribute-sentiment words. Mask the sentiment words and attribute-sentiment words.

[0014] S234. Optimize the sentiment encoding module based on the masked sentiment words and attribute-sentiment words;

[0015] S24. Input the user comment data into the sentiment encoding module to obtain the sentiment vector S. i ;

[0016] S3. Construct an emotion-behavior matching detection model; input the behavior vector and feature vector into the trained emotion-behavior matching detection model to obtain the matching degree W between the emotion vector and the behavior vector;

[0017] S31, Optimize generator G;

[0018] S32, Optimize discriminator D;

[0019] S33. Calculate the Wasserstein distance between the sentiment vector and the behavior vector using the optimized generator G and the optimized discriminator D.

[0020] S4. Construct a sequence graph neural network recommendation model based on user credibility; input user behavior data and user comment data into the sequence graph neural network recommendation model, calculate user activity UAI, combine W and UAI to obtain the user credibility quantification index T; predict user preferences based on T to obtain a recommendation list after weakening fake comments;

[0021] S41. Calculate user activity level UAI based on user behavior data and user comment data, and calculate user credibility level T based on UAI and W;

[0022] S42. Construct an adjacency matrix A∈R based on T and user-product interaction data. N×M , used to represent the interaction relationship between user and item;

[0023] S43. Perform graph convolution on A to obtain the updated embedding vector.

[0024] S431. Calculate the normalized weight of each neighbor node u relative to node v;

[0025] S432. Perform feature aggregation on the features of neighbor node u according to the normalized weights;

[0026] S433. Perform feature transformation on the aggregated features to obtain weighted features;

[0027] S434. Apply the activation function σ to the weighted features to obtain the feature representation of node v at layer (i+1). And use it as the updated embedding vector;

[0028] S44. Calculate the similarity between the embedded vectors to obtain the preference score of user u for item v. u,v ;

[0029] S45, based on score u,v Generate a recommendation list with fake reviews mitigated;

[0030] S5. Recommend products to the user based on the product recommendation list.

[0031] The beneficial effects of this invention include: by quantifying the degree of matching between user emotions and behaviors and using it as an indicator of user credibility, the influence of highly credible user behaviors on recommendation results is increased, thereby effectively weakening the influence of fake review data and improving the accuracy of e-commerce platform recommendation systems. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the e-commerce recommendation method based on fake review detection optimization in this invention;

[0033] Figure 2 This is a schematic diagram of the emotion encoding module in this invention;

[0034] Figure 3 This is a schematic diagram of the structure of the emotion-behavior matching detection model in this invention;

[0035] Figure 4 This is a schematic diagram of the sequence graph neural network recommendation model in this invention. Detailed Implementation

[0036] To make the objectives, technical solutions, features, and advantages of the present invention clearer, the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0037] This embodiment includes:

[0038] S1. Obtain user behavior data, user review data, and user-product interaction data from e-commerce platforms.

[0039] S2. Construct an emotion-behavior vectorization model; input user behavior data and user comment data into the emotion-behavior vectorization model to obtain behavior vectors and emotion vectors.

[0040] S21. Obtain the set of historical behavior sequence features of users in the user behavior data, B.

[0041] Specifically, the feature set B can be represented as:

[0042] B = {b1, b2, ... b} n ,ΔT}

[0043] ΔT={δt1,δt2,...,δt n |δt1=0,δt i =time(b i )-time(b i-1 )i>1}

[0044] Where ΔT represents the time range, which is a set of time intervals, B represents the set of user behavior sequence features within ΔT, and δt i Indicates the i-th action b i Compared with the previous action b i-1 The time interval between them, time(b) i ) represents user behavior b i The time of occurrence.

[0045] S22. Embed B to obtain the behavior vector representing user behavior.

[0046] Specifically, the formula for embedding representation is:

[0047] B emb =W b ·b i

[0048] Among them, W b It is an embedding matrix, B emb It is the embedding vector of the behavior, that is, the user's behavior vector.

[0049] S23. Construct and optimize the emotion coding module.

[0050] Specifically, the TF-IDF (Term Frequency–Inverse Document Frequency) method and the DBSCAN clustering method are used to obtain the sentiment seed word set; including the following steps:

[0051] S231. Replace user comment data with word sequences and process the word sequences to generate candidate words.

[0052] Specifically, when converting user comment data into word sequences, word segmentation and stop word removal are performed; the TF-IDF method is used to mine keywords from the word sequences. The formula for mining keywords using the TF-IDF method is as follows:

[0053] TF-IDF = TF × IDF

[0054] Where TF stands for Term Frequency, which represents the frequency with which a term appears in the text, and IDF stands for Inverse Document Frequency, which measures the importance of a term in a document.

[0055] Furthermore, sentiment words in the keywords are identified as candidate words based on a pre-defined sentiment dictionary.

[0056] S232. Cluster the candidate words using the DBSCAN clustering method to obtain the sentiment seed word set.

[0057] Specifically, the input to the DBSCAN clustering method includes: a sample set D = {d1, d2, ... dn}. m The DBSCAN clustering method outputs the following parameters: domain parameters (∈, MinPts), sample distance metric, and sentiment seed word set C = {C1, C2, ..., C}. K}

[0058] Furthermore, the DBSCAN clustering method processes the data in the following steps:

[0059] Step 1: Initialize the core object collection Initialize the number of clusters k = 0, initialize the set of unvisited samples Γ = D, and then perform cluster partitioning.

[0060] Step 2: For j = 1, 2, ..., m, find all core objects:

[0061] Find sample d using a distance metric. j The neighborhood ∈ - subsample set Nε(d j If the number of samples in the subset satisfies |Nε(d) j If |≥Minpts, then the sample x j Add to the core object sample set: Ω=Ω∪{d j}

[0062] Step 3, if the core object collection The algorithm ends if the condition is met; otherwise, proceed to step 4.

[0063] Step 4: In the core object set Ω, randomly select a core object O, initialize the current cluster core object queue Ωcur = {O}, initialize the category index k = k + 1, and initialize the current cluster sample set C. k ={O}, update the unvisited sample set Γ = Γ - {O}.

[0064] Step 5: If the current cluster core object queue Ωcur = {O}, then the current cluster C k Once generated, update the cluster partition C = {C1, C2, ... C}. K}, Update the core object set Ω = Ω - Ck Proceed to step 3.

[0065] Step 6: Take a core object O′ from the current cluster core object queue Ωcur, find all ∈-neighborhood subsets N∈(O′) using the neighborhood distance threshold ∈, let Δ=N∈(O′)∩Γ, and update the current cluster sample set C. k =C k ∪Δ, update the unvisited sample set Γ=Γ-Δ, update Ωcur=Ωcur∪(Δ∩Ω)-O′, go to step 5.

[0066] S233. Use a sentiment seed word set to identify sentiment words in user comment data, and further identify the attribute words that match the sentiment words. The attribute words and sentiment words together form attribute-sentiment words. Mask the sentiment words and attribute-sentiment words.

[0067] Specifically, dependency parsing is used when identifying attribute words that match sentiment words.

[0068] Furthermore, masking is performed based on sentiment words and attribute-sentiment words. Since sentiment words and attribute words are often key information carriers in text, masking these words can improve the model's generalization ability to unknown or unseen data. During masking, it is ensured that the number of masked tokens does not exceed 12% of the total number of tokens in the entire sentence; if it is less than 12%, other sentiment words in the sentence are masked until the proportion reaches 12%.

[0069] S234. Optimize the sentiment encoding module based on the sentiment words after masking and the attribute-sentiment words.

[0070] Specifically, the optimization function L is used to optimize the emotion encoding module. The formula for L is:

[0071] L = L sp +L sw +L E

[0072] Among them, L sp For optimizing the objective function of emotion polarity, L sw For the sentiment word objective optimization function, L E Optimize the function for the attribute-sentiment word objective.

[0073]

[0074]

[0075] Where, p i This indicates whether a word is a sentiment word; that is, if the word at position i is a word masked in the input sequence, then p... i =1, otherwise 0, Vi Given the input sequence, It is the polarity of hidden emotional words. for The probability estimate, W sp and b sp These are the trainable parameters in the emotion polarity objective optimization function. It is the original emotional word x i Based on the one-hot encoding representation of the entire vocabulary, For V i The probability distribution, W sw and b sw These are trainable parameters in the sentiment word objective optimization function, where m represents the number of masked attribute words in the input sequence. This is a sparse representation of attribute words. for The probability estimate, W E and b E These are the trainable parameters in the attribute objective optimization function.

[0076] S24. Input the user comment data into the sentiment encoding module to obtain the sentiment vector S. i .

[0077] Specifically, the input sequence V of the emotion encoding module i Character vector Z from user comment data i With position vector P i It was pieced together. (Z) i Z is obtained by word embedding of user comment data; the word sequence is then processed by the Word2Vec word embedding model. i P i The formula is:

[0078] P i (pos,2i)=sin(pos / 10000 2i / d )

[0079] P i (pos,2i+1)=cos(pos / 10000 2i / d )

[0080] Z i With P i The splicing method is as follows:

[0081] V i =P i +Z i

[0082] Where pos represents the position of a word with a word vector length of N, ranging from 0 to N; i represents the position of the smallest unit of the word vector, ranging from 0 to 253; and d represents the maximum length of the word vector, which is 254.

[0083] S3. Construct an emotion-behavior matching detection model; input the behavior vector and feature vector into the trained emotion-behavior matching detection model to obtain the matching degree W between the emotion vector and the behavior vector.

[0084] Specifically, an emotion-behavior matching detection model based on Wasserstein distance is used. The emotion-behavior matching detection model consists of a generator and a discriminator. The generator G is a projection function used to map emotion vectors and behavior vectors to the same feature space, using the emotion vector space as the target space and mapping the behavior vectors to the target space, thus achieving a unified space. The discriminator is used to calculate the Wasserstein distance between the emotion vectors and behavior vectors, and uses this distance as the degree of emotion-behavior matching. The steps include:

[0085] S31, Optimize generator G.

[0086] Specifically, the goal of generator G is to transform the behavior vector B emb Mapping this onto the emotional space results in the mapped behavioral vector B. emb ′=G(B emb It should be in the same space as the emotion vector as possible.

[0087]

[0088]

[0089] Among them, l proj Let G be the loss function of the generator. For the emotion vector S i and the mapped behavior vector G(B) emb The expected value of the Wasserstein distance of W(S) i ,G(B emb )) is the emotion vector S i and the mapped behavior vector G(B) emb The Wasserstein distance between them.

[0090] S32, Optimize the discriminator D.

[0091] Specifically, the goal of discriminator D is to quantify the degree of matching between the sentiment vector and the mapped behavior vector, i.e., the Wasserstein distance. The Wasserstein distance of fake reviewers is larger than that of genuine reviewers. The formula for optimizing discriminator D is:

[0092]

[0093]

[0094] Among them, l disc Let D be the loss function of the discriminator D, and D(S) be the loss function of the discriminator D. i ,G(B emb ) represents the discriminator D's relationship to the emotion vector and the mapped behavior vector G(B). emb The output of ) represents the matching score between the sentiment vector and the behavior vector.

[0095] S33. Calculate the Wasserstein distance between the sentiment vector and the behavior vector using the optimized generator G and the optimized discriminator D.

[0096] Specifically, the calculation formula is as follows:

[0097]

[0098] Where, γ∈(S i B emb ′) is the joint distribution of all possible samples, and d(p,q) is the distance metric between samples.

[0099] S4. Construct a sequence graph neural network recommendation model based on user credibility; input user behavior data and user comment data into the sequence graph neural network recommendation model, calculate user activity UAI, combine W and UAI to obtain the user credibility quantification index T; predict user preferences based on T to obtain the recommendation list after weakening fake comments.

[0100] S41. Calculate user activity level UAI based on user behavior data and user comment data, and calculate user credibility level T based on UAI and W.

[0101] Specifically, calculating User Activity (UAI) includes calculating User Interaction Score (UIS), Purchase Frequency (PF), Browse-to-Purchase Conversion Rate (BBCR), Cart Activity (CA), and Repeat Purchase Rate (RPR).

[0102] UAI=a·UIS+b·BBCR+c·CA+d·RPR+e·PF

[0103] UIS = Number of comments + Number of likes + Number of shares

[0104]

[0105]

[0106]

[0107]

[0108] Where a, b, c, d, and e represent the weights of user interaction rating, conversion rate from browsing to purchase, purchase activity, repeat purchase rate, and purchase frequency, respectively, and are allocated according to the contribution of each indicator to activity.

[0109] Furthermore, the formula for calculating user credibility T is:

[0110] T = μ·UAI + v·W(S) i B emb ′)

[0111] Where μ and v represent weights, W(S) i B emb ′) represents the degree of match between the commenter's sentiment and behavior, i.e., the Wasserstein distance W.

[0112] S42. Construct an adjacency matrix A∈R based on T and user-product interaction data. N×M , is used to represent the interaction relationship between a user and an item.

[0113] Specifically, the formula for constructing the adjacency matrix A is:

[0114] A′ uv =T=μ·UAI+v·W(S i B emb ′)

[0115] A∈R N×M

[0116] Where N is the number of users and M is the number of items. A′ uv It is the weight of one of the elements in the adjacency matrix A, representing the interaction strength between user node u and product node v.

[0117] S43. Perform graph convolution on A to obtain the updated embedding vector.

[0118] Furthermore, graph convolution operations include:

[0119] S431. Calculate the normalized weight of each neighbor node u relative to node v.

[0120] S432. Perform feature aggregation on the features of neighbor node u according to the normalized weights.

[0121] S433. Perform feature transformation on the aggregated features to obtain weighted features.

[0122] S434. Apply the activation function σ to the weighted features to obtain the feature representation of node v at layer (ι+1). And use it as the updated embedding vector.

[0123] Specifically, the formula for calculating graph convolution operations is:

[0124]

[0125] in, Let A' be the feature representation of node v at layer (ι+1), σ be the activation function, N(v) be the set of neighboring nodes u of node v, and A' be the feature representation of node v at layer (ι+1). uv′ Let be the weight of one of the elements in matrix A, representing the interaction strength between user node u and product node v′. W represents the feature representation of node u at layer ι. ι is the weight matrix of the ι-th layer, used to transform node features.

[0126] S44. Calculate the similarity between the embedded vectors to obtain the preference score of user u for item v. u,v .

[0127] Specifically, the similarity between embedding vectors is used to predict user preferences for items, and the calculation formula is as follows:

[0128]

[0129] in, Let be the feature representation of node u at layer ι. The feature representation of node v at layer l

[0130] S45, based on score u,v Generate a recommendation list with weakened fake reviews.

[0131] Specifically, for all items v that user u might be interested in, a preference score is calculated, and the K items with the highest scores are selected as recommendations:

[0132] Recommendations u =\topK({scre u,v |u∈χ})

[0133] Where χ represents the set of all items, and \topK represents selecting the K items with the highest scores.

[0134] S5. Recommend products to the user based on the product recommendation list.

[0135] Finally, it should be noted that the above description only depicts some embodiments of the present invention. For those skilled in the art, various changes, modifications, substitutions, and variations can be conceived of these embodiments without departing from the principles and spirit of the present invention. The scope of protection of the present invention is defined by the appended claims and their equivalents, and all the above-mentioned behaviors should be covered within the scope of protection of the present invention.

Claims

1. An e-commerce recommendation method based on fake review detection and optimization, characterized in that, The model's data processing includes: S1. Obtain user behavior data, user review data, and user-product interaction data from e-commerce platforms; S2. Construct a sentiment-behavior vectorization model; input user behavior data and user comment data into the sentiment-behavior vectorization model to obtain behavior vectors and sentiment vectors; S3. Construct an emotion-behavior matching detection model; input the behavior vector and feature vector into the trained emotion-behavior matching detection model to obtain the matching degree W between the emotion vector and the behavior vector; S4. Construct a sequence graph neural network recommendation model based on user credibility; input user behavior data and user review data into the sequence graph neural network recommendation model, calculate user activity UAI, combine W and UAI to obtain the user credibility quantification index T; predict user preferences based on T to obtain a product recommendation list after weakening fake reviews; The data processing of the sequence graph neural network recommendation model based on user credibility includes: S41. Calculate user activity level UAI based on user behavior data and user comment data, and calculate user credibility level T based on UAI and W; S42. Construct an adjacency matrix based on T and user-product interaction data. , used to represent the interaction relationship between user and item; S43, to Perform graph convolution to obtain the updated embedding vector. ; S44. Calculate the similarity between the embedded vectors to obtain the user's... For items Preference score ; S45, according to Generate a recommendation list with fake reviews mitigated; S5. Recommend products to the user based on the product recommendation list.

2. The e-commerce recommendation method based on fake review detection and optimization according to claim 1, characterized in that, The sentiment-behavior vectorization model processes user behavior data and user comment data in the following ways: S21. Obtain the set of historical behavior sequence features of users from user behavior data. ; S22, to By performing embedding representation, we obtain behavior vectors representing user actions. ; S23. Construct and optimize the emotion coding module; S24. Concatenate sequences based on user comment data. ,Will The sentiment encoding module is pre-trained to obtain sentiment vectors. .

3. The e-commerce recommendation method based on fake review detection and optimization according to claim 2, characterized in that, The construction of the emotion encoding module includes: S231. Replace user comment data with word sequences, and process the word sequences to select candidate words; S232. Cluster the candidate words using the DBSCAN clustering method to obtain the sentiment seed word set; S233. Use the sentiment seed word set to identify sentiment words in user comment data, and identify the matching attribute words based on the sentiment words. Combine the attribute words and sentiment words to form attribute-sentiment words, and mask the sentiment words and attribute-sentiment words. S234. Optimize the sentiment encoding module based on the sentiment words after masking and the attribute-sentiment words.

4. The e-commerce recommendation method based on fake review detection and optimization according to claim 3, characterized in that, The word sequence processing includes: using the TF-IDF method to mine keywords in the word sequence; and identifying sentiment words in the keywords as candidate words based on a pre-defined sentiment dictionary.

5. The e-commerce recommendation method based on fake review detection optimization according to claim 3, characterized in that, Dependency syntax is used to identify sentiment words and their matching attribute words.

6. The e-commerce recommendation method based on fake review detection optimization according to claim 1, characterized in that, The sentiment-behavior matching detection model consists of a generator and a discriminator. The generator maps sentiment vectors and behavior vectors to the same feature space, while the discriminator calculates the Wasserstein distance between sentiment vectors and behavior vectors.

7. The e-commerce recommendation method based on fake review detection optimization according to claim 1, characterized in that, Calculating User Activity (UAI) includes calculating User Interaction Score (UIS), Purchase Frequency (PF), Browse to Purchase Conversion Rate (BBCR), Purchase Activity (CA), and Repeat Purchase Rate (RPR).

8. The e-commerce recommendation method based on fake review detection optimization according to claim 1, characterized in that, right Graph convolution operations include: S431. Calculate the normalized weight of each neighbor node u relative to node v; S432. Perform feature aggregation on the features of neighbor node u according to the normalized weights; S433. Perform feature transformation on the aggregated features to obtain weighted features; S434, Activate the function Applying this to the weighted features yields the nodes. In the Layer feature representation And use it as the updated embedding vector.

Images