False marketing information propagation prediction method based on credibility and correlation degree

Abstract

The invention belongs to the field of e-commerce information dissemination, and relates to a false marketing information dissemination prediction method based on credibility and association degree, which comprises the following steps: acquiring user and false marketing information data and inputting the data into a trained dissemination prediction model to obtain a prediction result; the training process of the propagation prediction model comprises the following steps: inputting evaluation information of marketing information into a popularity and trust degree measurement model to obtain popularity Popl and trust degree Crel of the marketing information; inputting the user historical behaviors, the user social network and the marketing subject information into an association degree measurement model to obtain the forwarding activeness Act (ui), the interaction degree Int (ui, uj) and the association degree Rel (ui) with the marketing information of the user; inputting the Act (ui), the Int (ui, uj), the Rel (ui), the Popl and the Crel into an evolutionary game model to obtain influence Mutin (ui) and Mutout (ui) of user behavior selection; inputting the Mutin (ui) and the Mutout (ui) into the user behavior prediction model to obtain a prediction result; updating the propagation prediction model according to the prediction result until a trained propagation prediction model is obtained; according to the method, the user behavior tendency is quantified according to the credibility and the correlation degree by utilizing an evolutionary game, so that more accurate prediction can be realized.

Description

Technical Field

[0001] The present invention belongs to the field of e-commerce information dissemination, and in particular relates to a method for predicting the dissemination of false marketing information based on trust and relevance. Background Art

[0002] With the increasing popularity of the internet and the rapid development of online shopping, some online merchants are engaging in false advertising to attract users and increase sales. The spread of false marketing information has become a serious challenge and a source of significant concern. False marketing information spreads rapidly through various channels, disrupting market order and potentially having a serious negative impact on consumers, undermining their interests and trust in online shopping. In light of this phenomenon, studying the spread of false marketing information not only helps regulate the market environment and protect consumer interests, but also provides a basis for formulating sound policies and regulations, promoting the healthy development of the market economy. By thoroughly analyzing the patterns of false information dissemination, relevant departments, businesses, and consumers can more effectively prevent and address this social issue, enhancing overall social integrity and market competitiveness.

[0003] Currently, research on the spread of false marketing information focuses on two main areas: one area draws on principles of epidemiology to classify the node states of users in social networks and construct state transition equations to explore the rumor propagation process. The other area primarily uses machine learning or deep learning techniques, incorporating factors influencing dissemination to predict user forwarding behavior.

[0004] Users' trust and relevance in false marketing information play a significant role in its spread, influencing their cognition and behavioral decisions. Whether users spread unverified false marketing information is often influenced by a combination of these two factors, and the effects vary. Therefore, understanding how trust and relevance influence the spread of false marketing information and effectively predicting user behavior are crucial for fostering a healthy and reliable marketing market. To achieve prediction of false marketing information spread based on trust and relevance, the following challenges remain in the research process:

[0005] 1. The impact of product reviews on user trust is difficult to measure. Users' trust in false marketing information is influenced by product reviews, and positive and negative reviews have different impacts on user trust. Therefore, measuring the sentiment in reviews and the different impacts of these two types of reviews is a challenge.

[0006] 2. The correlation between users and false marketing information is difficult to measure. Assessing the correlation between users and false marketing information involves complex relationship data between users and products, as well as user entities. Mining this data and using it to quantify the correlation is a challenge.

[0007] 3. Trust and relevance have different effects on the spread of false marketing information. Trust and relevance have different impacts on user behavior, making quantifying this effect and predicting user behavior a challenge. Summary of the Invention

[0008] To solve the above-mentioned problems in the prior art, the present invention adopts a flowchart of a method for predicting the spread of false marketing information based on trust and relevance, comprising: obtaining e-commerce platform users and false marketing information data, inputting the e-commerce platform users and false marketing information data into a trained false marketing information spread prediction model to obtain a prediction result; the false marketing information spread prediction model includes: a popularity and trust measurement model, a relevance measurement model, an evolutionary game model, and a prediction model;

[0009] The training process of the false marketing information propagation prediction model includes:

[0010] S1. Obtain a dataset of e-commerce platform users and false marketing information, including user historical behavior data, user social network data, marketing subject information of marketing information, and evaluation information;

[0011] S2. Inputting the evaluation information of the marketing information into the popularity and trust measurement model to obtain the popularity and trust of the marketing information;

[0012] S3. Inputting user historical behavior data, user social network data, and marketing subject information of marketing information into a correlation measurement model to obtain user forwarding activity, user interaction, and correlation between users and marketing information;

[0013] S4. Input user forwarding activity, user interaction, user-marketing information relevance, marketing information popularity, and marketing information trust into the evolutionary game model to obtain the influence of internal and external factors;

[0014] S5. Inputting user forwarding activity, user interaction, user-marketing information relevance, marketing information popularity, marketing information trust, and the influence of internal and external factors into a user behavior prediction model to obtain a user behavior prediction result;

[0015] S6. Calculate the loss function value based on the user behavior prediction results, and update the parameters of the false marketing information propagation prediction model based on the loss function value. When the loss function value is minimized, a trained false marketing information propagation prediction model is obtained.

[0016] Beneficial effects:

[0017] 1. The present invention distinguishes between positive and negative reviews through a classifier and introduces LSTM to model the intensity of emotions. It not only takes into account the singleness of evaluation emotions, but also further quantifies the impact of positive and negative emotions on user trust, so that the impact of evaluation information on user trust is more finely modeled, and the accuracy of the model in processing user trust is enhanced; 2. The present invention adopts the User2vec vector representation method to map user historical behavior data, user social network data, and marketing subject information of marketing information into a low-dimensional space, and then uses the knowledge representation method to calculate user forwarding activity, user interaction, and the correlation between users and marketing information, revealing the complex relationship between users and products, and enhancing the interpretability and accuracy of the model in practical applications; 3. The present invention uses evolutionary game to quantitatively study user behavior tendencies based on the trust of marketing information and the correlation between users and marketing information, and obtains the influence of trust and correlation on user behavior choices. It also uses the advantages of GCN in social network data processing to predict the influence of trust and correlation on user behavior choices, thereby more accurately predicting users' reactions to false marketing information and the evolution of behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of a method for predicting the spread of false marketing information based on trust and relevance provided by an embodiment of the present invention;

[0019] Figure 2 A schematic diagram of the classification of evaluation information provided by an embodiment of the present invention;

[0020] Figure 3 A flowchart of a method for predicting the spread of false marketing information based on trust and relevance provided by an embodiment of the present invention; DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, the present invention adopts a false marketing information propagation prediction method based on trust and relevance, including:

[0023] S1. Obtain user historical behavior data, user social network data, marketing subject information of marketing information, and evaluation information;

[0024] In this embodiment, user historical behavior includes the user sending or receiving marketing information; user social network data is data of users interacting with other users in various ways, and the interaction methods between users include sending and receiving messages, commenting and replying, and liking and forwarding; the marketing subject information of the marketing information includes: descriptive information of the marketing subject (product); the evaluation information of the marketing information is the evaluation information of the product corresponding to the marketing information, including the evaluation text and its evaluation time.

[0025] Methods for obtaining user historical behavior data, user social network data, marketing information, and evaluation information can include obtaining raw data from data sources provided by e-commerce platforms or directly downloading existing public data sources. This raw data is typically unstructured and cannot be directly used for data analysis. Simple data cleaning can structure most unstructured data. For example, deleting duplicate data and cleaning invalid nodes can be used.

[0026] S2. Inputting the evaluation information of the marketing information into the popularity and trust measurement model to obtain the popularity and trust of the marketing information;

[0027] like Figure 2 As shown, the popularity and trust measurement model processes the evaluation information of marketing information including:

[0028] S21. Calculate the weight vector of each evaluation information of the marketing information using the term frequency-inverse document frequency representation method;

[0029] The weight vector for each evaluation information of the marketing information is calculated as follows:

[0030] S211, evaluation information set D = {d1, d2, ..., d n}, each element in the set D represents a piece of evaluation information, where d j represents the jth evaluation information, and n represents the number of evaluation information; by performing text preprocessing and word segmentation operations on the evaluation information in the set D, a term table T containing all unique terms is constructed. m}, each element in the set T represents a term, t i represents the i-th term, and m represents the number of terms;

[0031] S212, calculate the word frequency tf of each word in each evaluation information ij :

[0032]

[0033] Among them, n i,j For the entry t i In the evaluation information j The number of times it appears in k n k,j For evaluation information j The number of all entries in , k is the evaluation information d j Index of entries in ;

[0034] S213, calculate the inverse document frequency idf of each term i :

[0035]

[0036] Among them, |D| is the total number of evaluation information in the current evaluation information set, |{j:t i ∈d j}| indicates that the term t is included i The number of evaluation information (i.e. n i,j ≠0), if the word is not in the current evaluation information d j In the case of , the denominator will be zero, so use |{j:t i ∈d j}|+1.

[0037] S214, calculate the weight vector {η of each evaluation information according to the word frequency and reverse file frequency of the entry j,1 ,η j,2 ,...,η j,n}, η j,i Indicates the term t i In d j The weight in is calculated as follows:

[0038] η j,i =tf i,j ×idf i

[0039] S22. Use the naive Bayes classifier to classify each evaluation information according to the weight vector of the evaluation information to obtain the positive evaluation information set D pos and negative evaluation information set D neg ;

[0040] Positive review set D pos ={d1,d2,...,d N}, negative review set D neg ={d1,d2,...,d M}; where N is the number of reviews in the positive review set, and M is the number of reviews in the negative review set; since each weight vector can be considered statistically independent, the naive Bayes classifier can handle these vectors well and perform classification based on the vector features.

[0041] S23, based on the positive evaluation information set D pos and negative evaluation information set D neg Calculate the credibility of marketing information;

[0042] Calculating the credibility of marketing information includes:

[0043] S231, respectively, the positive evaluation information set D pos and negative evaluation information set D neg The evaluation information is input into the LSTM model for modeling, and the positive evaluation information vector set d is obtained. pos and negative evaluation information vector set d neg ;

[0044] S232. Calculate the weight λ of the positive and negative evaluation information on the trustworthiness based on the positive evaluation information vector and the negative evaluation information vector. n ;

[0045] Specifically, according to the positive evaluation information vector set d pos and negative evaluation information vector set d neg Calculate the emotional intensity of positive evaluation information and negative evaluation information separately:

[0046] E oos =s pos *d pos

[0047] E neg =s pos *d neg

[0048] Among them, s pos and s neg A learnable weight matrix representing the emotions of positive and negative evaluation information, respectively. The weight matrix includes the scores of emotional words of different intensities, and the score ranges from 0 to 1.

[0049] The weights λ of the impact of positive and negative reviews on trust are calculated based on the emotional intensity of positive and negative reviews. n :

[0050]

[0051] S233, based on the positive evaluation information set D pos , negative evaluation information set D negand weight λ n Calculate the confidence Cre:

[0052] Cre=Q pos +λ n Q neg

[0053]

[0054] Among them, Num pos Represents the positive evaluation information set D pos The number of positive reviews in neg Represents the negative evaluation information set D neg Num[All] represents the total number of negative reviews of the current marketing information, Q pos , Q neg The number Num pos and number Num neg The ratio of the total number of evaluation messages to marketing messages.

[0055] S24. Calculate the popularity of the marketing information based on the evaluation information of the marketing information.

[0056] Information popularity describes the prevalence of fake marketing information. On social networks, information spreads rapidly, and its popularity often peaks shortly after dissemination and then begins to decline. This phenomenon is similar to the half-life of an element, so information popularity is defined using the half-life function as follows:

[0057]

[0058] Among them, Pop0 represents the peak value of information popularity, t and t0 represent the current time and the time when the evaluation information of marketing information first appears, respectively, and T 1 / 2 Represents the half-life of popularity.

[0059] S3. Inputting user historical behavior data, user social network data, and marketing subject information of marketing information into a correlation measurement model to obtain user forwarding activity, user interaction, and correlation between the user and marketing information;

[0060] like Figure 3 As shown in Figure 2, the association measurement model processes user historical behavior data, user social network data, and marketing subject information in the following ways:

[0061] S31. Mapping user historical behavior data, user social network data, and marketing subject information into a low-dimensional vector space using the User2vec vector representation method to obtain low-dimensional user historical behavior data, user social network data, and marketing subject information; wherein User2vec is user vectorization;

[0062] S32. Based on low-dimensional user historical behavior data, user social network data, and marketing subject information, a knowledge representation method is used to describe the complex and diverse relationships between users and marketing information, thereby obtaining user forwarding activity, user interaction, and the correlation between users and marketing information.

[0063] Specifically, calculating user forwarding activity, user interaction, and the correlation between users and marketing information includes:

[0064] S321, calculate the user forwarding activity Act (u i );

[0065] Specifically, the knowledge representation method is used to calculate each user u based on the low-dimensional user historical behavior data. i The number of marketing messages forwarded in history and the number of marketing messages received by each user in history are used to calculate the user forwarding activity Act(u i ):

[0066]

[0067] Among them, transNum(u i ) represents user u i The number of marketing information forwarded in history, allNum(u i ) represents the number of marketing messages the user has received in the past;

[0068] S322, using the knowledge representation method to calculate the interest matching degree Mat(e) between the user and the marketing subject based on the low-dimensional user historical behavior data and marketing subject information l ,u i );

[0069] Specifically, the knowledge representation method is used to calculate the similarity between each keyword and all other keywords in the marketing subject information (i.e., product description information) of low-dimensional marketing information, so as to find the relationship between keywords and extract the most relevant keyword topic (e.g., l ) (i.e. the keyword with the highest similarity); extract the low-dimensional user historical behavior data through Word2Vec to obtain the user interest preference label Inter(u i ); marketing subject according to marketing information iThe most relevant keywords and user interest preference tags are used to calculate the interest matching degree between the user and the marketing subject Mat(e l ,u i ), defined as:

[0070]

[0071] Among them, e l For the first marketing information MI l The marketing subject is the product.

[0072] S323, calculate the interaction degree Int(u) between users based on the low-dimensional user social network data i ,u j );

[0073] Specifically, the knowledge representation method is used to calculate the number of interactions Num[Interact k (u i ,u j )], each user u i The number of interactions with all other users in each mode k Num[Interact k (u i )], and then calculate the interaction degree Int(u i ,u j ):

[0074]

[0075] Among them, λ k Indicates the weight of the k-th interaction mode, Num[Interact k (u j )] represents user u j The total number of interactions with all users in the kth way;

[0076] S324, calculate the correlation between the user and the marketing information Rel (e) based on the interest matching degree between the user and the marketing subject and the user's interaction degree. l ,u i ); defined as follows:

[0077]

[0078] Among them, Mat(e l ,u j ) represents user u j With marketing entities l Interest matching degree, Int(u i ,u j ) represents user ui With user u j The correlation between m Represents the learnable weight of the relationship between the user and the current marketing entity and the correlation between users and the correlation between users and false marketing information.

[0079] S4. Input the trust of marketing information and the relevance between users and marketing information into the evolutionary game model to obtain the influence of internal and external factors;

[0080] The specific steps include:

[0081] S41. Calculate the impact factors of internal and external factors:

[0082] fac in (u i ,MI l )=Act(u i )×Rel(el,u i )

[0083] fac out (u i ,MI l )=Pop l ×Cre l

[0084] Among them, fac in (u i ,MI l ) and fac out (u i ,MI l ) represent the influencing factors of internal factors and external factors respectively. Internal factors include user forwarding activity Act(u i ) and the correlation between users and marketing information Rel(e l ,u i ), external factors include information popularity Pop l and information trust Cre l , MI l This is the first marketing message.

[0085] S42. Conduct evolutionary games on the influencing factors of internal factors and external factors respectively to obtain the influence of internal factors and external factors on user behavior choices;

[0086] The specific formula is:

[0087]

[0088] Mut in (u i )={Mut in (ui ,MI1),Mut in (u i ,MI2),…,Mut in (u i ,MI L )}

[0089] Mut out (u i )={Mut out (u i ,MI1),MI out (u i ,MI2),…,Mut out (u i ,MI L )}

[0090] Among them, Mut in (u i ,MI l ) and Mut in (u i ,MI l ) represent the influence of internal factors and external factors on user behavior choices after quantification through evolutionary game, Mut in (u i ) and Mut in (u i ) represent the influence of all internal and external factors on user behavior choices after quantification through evolutionary game.

[0091] S5. Inputting user forwarding activity, user interaction, user-marketing information relevance, marketing information popularity, marketing information trust, and the influence of internal and external factors into a user behavior prediction model to obtain a user behavior prediction result;

[0092] Graph structure data is constructed based on the input data, and the graph structure data is input into the user behavior prediction model to obtain the user behavior prediction results; the user behavior prediction model is a graph neural network model; user behavior prediction is a binary classification problem, that is, whether the user will participate in the spread of false marketing information.

[0093] The nodes of the graph structure are users and marketing information; each user node u i The characteristics include: user forwarding activity Act(u i ), Internal factors influence Mut in (u i ) and external factors influence Mut in (u i), these parameters respectively represent the degree of comprehensive influence of users by their own internal factors (forwarding activity and relevance) and external factors (information popularity and trust); each marketing information node MI l The characteristics include marketing information popularity Pop l and marketing trust Cre l The weight of the interaction edge between users is the user interaction degree Int(u i ,u j ), reflecting the intensity of social interaction; the weight of the association edge between the user and marketing information is the correlation degree between the user and information Rel(e l ,u i ).

[0094] During the information dissemination process, graph neural networks aggregate node features and edge weights to ultimately predict whether users participate in the dissemination of false marketing information.

[0095] In one embodiment, the user behavior prediction model is a graph attention network model; wherein, when calculating the attention coefficient between the user node i and the marketing information node l, the graph attention network uses the attention mechanism to dynamically adjust the feature Mut of the user node i. in (u i ,MI l ) and the feature Mut in (u i ,MI l ) is used to quantify the user's behavioral tendency towards specific marketing information.

[0096] S6. Calculate the loss function value based on the user behavior prediction results, and update the model based on the loss function value. When the loss function value is minimized, the trained false marketing information propagation prediction model is obtained. The loss function uses binary cross entropy loss:

[0097]

[0098] Where n is the number of users, y i is the true label of the i-th user (i.e., spreading marketing information or not spreading marketing information), is the predicted probability of the i-th user.

[0099] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation plans of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting the spread of false marketing information based on trust and relevance, characterized in that: include: Obtain e-commerce platform user and false marketing information data, and input the trained false marketing information propagation prediction model to obtain user behavior prediction results; The false marketing information propagation prediction model includes: popularity and trust measurement model, relevance measurement model, evolutionary game model and prediction model; The training process of the false marketing information propagation prediction model includes: S1. Obtain a dataset of e-commerce platform users and false marketing information, including user historical behavior data, user social network data, marketing subject information of marketing information, and evaluation information; S2. Inputting the evaluation information of the marketing information into the popularity and trust measurement model to obtain the popularity and trust of the marketing information; S3. Inputting user historical behavior data, user social network data, and marketing subject information of marketing information into a correlation measurement model to obtain user forwarding activity, user interaction, and correlation between users and marketing information; S4. Input user forwarding activity, user interaction, user-marketing information relevance, marketing information popularity, and marketing information trust into the evolutionary game model to obtain the influence of internal and external factors; S5. Inputting user forwarding activity, user interaction, user-marketing information relevance, marketing information popularity, marketing information trust, and the influence of internal and external factors into a user behavior prediction model to obtain a user behavior prediction result; S6. Calculate the loss function value based on the user behavior prediction results, and update the parameters of the false marketing information propagation prediction model based on the loss function value. When the loss function value is minimized, a trained false marketing information propagation prediction model is obtained.

2. The method for predicting the spread of false marketing information based on trust and relevance according to claim 1, characterized in that: The trust measurement model processes the evaluation information of marketing information including: S21, using the word frequency-inverse document frequency representation method to calculate the weight vector of each evaluation information; S22, using a classifier to classify each piece of evaluation information according to the weight vector of the evaluation information, to obtain a positive evaluation information set D pos and negative evaluation information set D neg ; S23, based on the positive evaluation information set D pos and negative evaluation information set D neg Calculate the credibility of marketing information; S24. Calculate the popularity of the marketing information based on the evaluation information of the marketing information.

3. The method for predicting the spread of false marketing information based on trust and relevance according to claim 2, characterized in that: According to the positive evaluation information set D pos and negative evaluation information set D neg Calculating the credibility of marketing information includes: S231, use LSTM model to analyze the positive evaluation information set D pos and negative evaluation information set D neg The evaluation information is processed to obtain a positive evaluation information vector set and a negative evaluation information vector set; S232, calculating the weight λ of the impact of positive and negative evaluation information on trust based on the positive evaluation information vector set and the negative evaluation information vector set n ; S233, based on the positive evaluation information set D pos , negative evaluation information set D neg and weight λ n Calculate the credibility of marketing messages.

4. The method for predicting the spread of false marketing information based on trust and relevance according to claim 3, characterized in that: Calculating the weights of the impact of positive and negative evaluation information on trust includes: calculating the emotional intensity of the positive evaluation information and the emotional intensity of the negative evaluation information based on the positive evaluation information vector set and the negative evaluation information vector set respectively, calculating the ratio of the emotional intensity of the positive evaluation information to the emotional intensity of the negative evaluation information, and obtaining the weights of the impact of positive and negative evaluations on trust.

5. The method for predicting the spread of false marketing information based on trust and relevance according to claim 3, characterized in that: Calculating the trustworthiness of marketing information includes: calculating the number of evaluation information Num in the positive evaluation information set of marketing information respectively pos And the number of evaluation information in the negative evaluation information set Num neg , calculate the number Num pos and number Num neg The ratio Q of the total number of evaluation information to marketing information pos , Q neg ; According to the weight λ n Contrast value Q pos , Q neg Perform weighted combination to obtain the credibility of marketing information.

6. The method for predicting the spread of false marketing information based on trust and relevance according to claim 1, characterized in that: The association measurement model processes user historical behavior data, user social network data, and marketing subject information in the following ways: S31, mapping the user's historical behavior data, the user's social network data, and the marketing subject information into a low-dimensional vector space to obtain the low-dimensional user's historical behavior data, the user's social network data, and the marketing subject information; S32. Calculate user forwarding activity, user interaction, and the correlation between the user and marketing information based on low-dimensional user historical behavior data, user social network data, and marketing entity information.

7. The method for predicting the spread of false marketing information based on trust and relevance according to claim 6, characterized in that: Calculating user forwarding activity, user interaction, and user relevance to marketing information includes: S321. Calculate user forwarding activity based on low-dimensional user historical behavior data; S322. Calculate the interest matching degree between the user and the marketing subject based on the low-dimensional user historical behavior data and marketing subject information; S323. Calculating user interaction degree based on low-dimensional user social network data; S324. Calculate the correlation between the user and the marketing information based on the interest matching degree between the user and the marketing entity and the user interaction degree.

8. The method for predicting the spread of false marketing information based on trust and relevance according to claim 1, characterized in that: The evolutionary game model processes user forwarding activity, user interaction, user-marketing information relevance, marketing information popularity, and marketing information trust, including: S41. Calculate the impact factor of internal factors based on the user's forwarding activity and the degree of association between the user and the marketing information, and calculate the impact factor of external factors based on the popularity and trustworthiness of the marketing information; S42. Conduct evolutionary games on the influencing factors of internal factors and external factors respectively to obtain the influence of internal factors and the influence of external factors.

9. The method for predicting the spread of false marketing information based on trust and relevance according to claim 8, characterized in that: The evolutionary game of the influencing factors of internal and external factors includes: Among them, Mut in (u i ,MI l ), Mut in (u i ,MI l ) represent the influence of internal factors and external factors after evolutionary game, fac in (u i ,MI l ), fac out (u i ,MI l ) represent the influencing factors of internal factors and external factors, MI l For the lth marketing information, u i For the i-th user.

10. The method for predicting the spread of false marketing information based on trust and relevance according to claim 1, characterized in that: The user behavior prediction model is a graph neural network model; The user behavior prediction model processes input data by: constructing graph structure data based on user forwarding activity, user interaction, the correlation between users and marketing information, the popularity of marketing information, the trust in marketing information, and the influence of internal and external factors, and inputting the graph structure data into the user behavior prediction model to obtain user behavior prediction results; wherein, the nodes of the graph structure data are users and marketing information; the characteristics of each user node include: user forwarding activity, the influence of internal factors and the influence of external factors; the characteristics of each marketing information node include: the popularity of marketing information and the trust in marketing information; the weight of the edge between users is the user interaction, and the weight of the edge between users and marketing information is the correlation between users and marketing information.

Images