Data processing method based on artificial intelligence

Abstract

The invention discloses a data processing method based on artificial intelligence. The method comprises the steps of obtaining an initial tendency score value of a user to a target business tweet; constructing network equipment feature information; according to the network equipment feature information and the initial tendency score value, a preset tendency determination algorithm is utilized to obtain a target tendency score value; according to the target tendency score values of all the target users, screening out potential users and differentiating service interaction modes of the potential users, sending service interaction information corresponding to the service interaction modes to the potential users, and obtaining feedback information of the potential users; and inputting the feedback information of the potential user into a pre-trained business type identification model to obtain a tendency business type of the potential user. Therefore, the response rate of the service interaction information is effectively improved, and the accuracy and effectiveness of target service tweet value evaluation are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a data processing method based on artificial intelligence. Background Art

[0002] To attract more business, service-oriented companies generally choose to promote their businesses through advertising tweets. For example, in the field of intellectual property, many agencies transform their professional and technical knowledge into tweets and send them to users for learning, while also forming a potential business advertisement. However, the current system cannot effectively evaluate the effectiveness of such business advertising tweets. Therefore, it is unable to provide users with effective reference information for adjusting and optimizing business tweets, resulting in the business tweets failing to maximize their economic value.

[0003] With the popularity and development of social media, user interactions on social media are increasing. Accurately assessing users' true inclinations towards business tweets has become a critical issue. Traditional assessment methods rely primarily on explicit interaction data, such as likes and comments, but this data often fails to fully reflect users' true intentions and behavioral habits. Furthermore, the significant differences in terminal device performance and network conditions among users pose challenges to accurately assessing user inclinations. Summary of the Invention

[0004] This application provides an artificial intelligence-based data processing method to avoid the waste of business interaction resources of target business tweets, effectively improve the response rate of business interaction information, and improve the accuracy and effectiveness of the value assessment of target business tweets.

[0005] This application provides an artificial intelligence-based data processing method, including: S101, collecting target user's interactive behavior data on target business tweets to obtain the target user's initial propensity score for the target business tweets; S102, obtaining network status parameters, terminal hardware information, and activity values ​​of target user interaction behavior data, and constructing network device feature information; S103, obtaining a target propensity score value using a preset propensity determination algorithm based on the network device feature information and the initial propensity score value; S104, based on the target propensity scores of all target users, screening potential users and differentiating the business interaction methods of the potential users, sending business interaction information corresponding to the business interaction methods to the potential users, and obtaining feedback information from the potential users; S105, inputting the potential user's feedback information into a pre-trained business type recognition model to obtain the potential user's preferred business type; S106, repeat steps S101 to S105 until the preset cycle time is reached, count the preferred business types of all potential users to form an actual feedback business library, compare the target proportion of business types in the target business tweets with the actual proportion of business types in the actual feedback business library, and obtain the push value of each business type in the target business tweets.

[0006] Preferably, the target users are all users who browse the target business tweets. The interactive behavior data includes the length of stay, page scroll depth, and like effect duration. The initial propensity score value is calculated according to the following formula:

[0007] Among them, S is the initial propensity score value, T is the length of stay, and D is the page scroll depth. is the maximum page scroll depth, The duration of the like effect. 、 and are the weight factors of the initial propensity score value, namely, the length of stay, the page scroll depth and the time of the like.

[0008] Preferably, the constructing of network device feature information specifically includes: A1. Obtain network status parameters and terminal hardware information, and use the pre-trained device network scoring model to obtain the device network score. A2. The device network score and activity value are combined to form network device feature information.

[0009] Preferably, the preset tendency determination algorithm includes: B1. Perform a weighted sum of the device network score and activity value in the network device feature information to obtain an interaction credibility value; B2. If the interaction credibility value is lower than the preset threshold, execute step B3; if the interaction credibility value is higher than or equal to the preset threshold, use the initial propensity score value as the target propensity score value; B3. Combining the initial propensity score value and network device feature information to form a multivariate effect feature vector for the user's recommendation of the target service, and inputting it into a pre-trained propensity score correction model to obtain a correction guidance value; B4. Use the correction guide value to correct the initial propensity score value to obtain the target propensity score value.

[0010] Preferably, the propensity score correction model is obtained as follows: C1. Collect network device feature information and initial propensity score values ​​of different target users whose interaction credibility values ​​for different business tweets are lower than the preset threshold in history, and form multiple historical multivariate effect feature vectors. C2. Label each historical multivariate effect feature vector, and set the label content as the correction guide value; C3. Use the labeled historical multivariate effect feature vectors as a training set, use the training set to train and learn the pre-selected neural network structure, optimize the model parameters, and obtain the propensity score correction model.

[0011] Preferably, the correction guide value is obtained in the following manner: D1. Based on each historical multivariate effect feature vector, obtain the immediate interest level of the corresponding user's feedback information; D2. Determine a correction guide value for the initial propensity score based on the initial propensity score value and its corresponding propensity threshold, the immediate interest level and its corresponding interest threshold.

[0012] Preferably, step D2 specifically includes: When the deviation direction between the initial propensity score and the corresponding propensity threshold is the same as the deviation direction between the immediate interest and the corresponding interest threshold, a correction guide value is obtained according to the first preset algorithm:

[0013] Among them, C is the correction guidance value, I is the immediate interest value, is the interest threshold, S is the initial propensity score value, is the tendency threshold, is a sign function that maintains the original deviation direction. k1 and k2 are pre-set weight values ​​used to adjust the influence of immediate interest and initial propensity score on the correction guidance value. k1 is greater than k2.

[0014] Preferably, the D2 further comprises: When the deviation direction between the initial propensity score and the corresponding propensity threshold is opposite to the deviation direction between the immediate interest and the corresponding interest threshold, the correction guidance value is obtained according to the second preset algorithm:

[0015] Among them, C is the correction guidance value, I is the immediate interest value, is the interest threshold, S is the initial propensity score value, is the tendency threshold, is a sign function that maintains the original deviation direction, and k3 and k4 are pre-set weight values.

[0016] Preferably, the S104 specifically includes: S201, matching corresponding interaction levels based on the target propensity scores of potential users based on pre-set score intervals, where each score interval corresponds to an interaction level; S202, presetting one or more business interaction modes for each interaction level; S203: Select a corresponding business interaction method according to the interaction level of the potential user, and send business interaction information to the potential user.

[0017] Preferably, the interaction level corresponds to the scoring interval one by one, and the scoring interval includes [ 、[ and[ , corresponding to low level, intermediate level and high level respectively, a is smaller than b and is preset according to actual situation.

[0018] One or more technical solutions provided in this application have at least the following technical effects or advantages: By introducing diversified interactive behavior data and network device feature information, it can more comprehensively reflect the user's interest in the target business tweets. Accurate user propensity assessment provides strong support for the subsequent mining of potential users of the target business tweets, avoiding the waste of business interaction resources for the target business tweets, effectively improving the response rate of business interaction information, and also improving the accuracy and effectiveness of the value assessment of the target business tweets. By considering the influence of network devices and status, it can maintain a high evaluation accuracy in different devices and network environments, avoiding deviations in the propensity score value.

[0019] By introducing interactive credibility values ​​and correction guidance values, the initial propensity score values ​​are corrected in a more detailed and reasonable manner, taking into account the impact of different device networks and user activity on user propensity, making the evaluation results more accurate; through the propensity score correction model, the deviation of the initial propensity score value can be automatically adjusted, which improves the intelligence and automation level of the evaluation and can better adapt to changes in different users and scenarios; by considering the immediate interest level and the comparison results of the initial propensity score value with the preset threshold, the initial propensity score value can be corrected in a targeted manner to obtain a more accurate target propensity score value.

[0020] Target users whose target propensity score values ​​are greater than the preset propensity threshold are regarded as potential users, and their corresponding interaction levels are determined based on the target propensity score values. Each interaction level corresponds to a business interaction method, so as to differentiate the business interaction methods (emotion, form) of different potential users, improve the effectiveness of potential user feedback information, and guide potential users to respond to more valuable questions or information, which is convenient for subsequent identification and judgment of business types. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Schematic diagram of the flow of the data processing method based on artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] To facilitate understanding of the present invention, the present application will be described more comprehensively below with reference to the relevant drawings; the drawings show preferred embodiments of the present invention, but the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to enable a more thorough and comprehensive understanding of the disclosed content of the present invention.

[0023] It should be noted that the terms “vertical”, “horizontal”, “up”, “down”, “left”, “right” and similar expressions used in this document are for illustrative purposes only and do not represent the only implementation method.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains; the terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention; the term "and / or" used herein includes any and all combinations of one or more of the associated listed items.

[0025] Example 1: Figure 1 It is a flowchart of the data processing method based on artificial intelligence according to an embodiment of the present invention.

[0026] When users browse business tweets on social media, they rely on their explicit interaction data such as likes and comments, but ignore the intention information contained in their implicit interaction data such as browsing time and page scrolling depth. In addition, the terminal device performance and network conditions of different users vary greatly, resulting in inconsistent tendencies and intentions contained in the same interactive behavior. For example, on low-performance devices or in a weak network environment, the user's browsing time may not fully reflect their interest in the content. Therefore, there is an urgent need for multi-dimensional data that can comprehensively utilize users' explicit and implicit interaction data and consider the influence of device and network factors to accurately assess users' true inclinations towards specific business content and provide strong support for subsequent personalized content push.

[0027] like Figure 1 As shown, a data processing method based on artificial intelligence includes the following steps: S101, collecting target user's interactive behavior data on target business tweets, and obtaining the target user's initial propensity score value for the target business tweets.

[0028] Target users refer to all users who browse the target business tweets. Interaction behavior data includes but is not limited to dwell time, page scroll depth, and like time. The initial propensity score is calculated according to the following formula:

[0029] Where S is the initial propensity score value, T is the dwell time, and D is the page scroll depth (unit: pixel or percentage). The maximum page scroll depth (according to the user terminal page settings), The duration of the like effect (calculated from the time the page is loaded to the time the like is clicked) takes into account the relative position of the like time within the duration of stay. The square term is used to emphasize the impact of the early or late like time on the score: if the user clicks the like button in the middle or late stages of the stay, it means that the user is more focused on and recognizes the content of the tweet, which can avoid users blindly clicking the like button without studying the content of the tweet. 、 and They are used to represent the influence of dwell time, page scrolling depth and like time on the initial propensity score value, and are preset based on actual conditions and expert experience.

[0030] S102, obtaining network status parameters, terminal hardware information and activity value of interactive behavior data generated by the target user, and constructing network device feature information.

[0031] Among them, network status parameters include but are not limited to network delay, packet loss rate and bandwidth, terminal hardware information includes but is not limited to device model, operating system version and CPU performance, and the activity value is set to the ratio of the user's total online time within the preset historical time window to the length of the preset historical time window. The preset historical time window is set according to actual conditions, for example, it is set to 5 minutes.

[0032] Specifically, constructing network device feature information includes: A1. Obtain network status parameters and terminal hardware information, and use the pre-trained device network scoring model to obtain the device network score.

[0033] Among them, the device network scoring model uses a large number of user-side network status parameters and terminal hardware information collected at historical moments as training data and labels them. The label annotation content is set to the device network score at the corresponding historical moment (which can be manually labeled or calculated using a related evaluation algorithm, which is not described in detail in the present invention). The device network score is a value between 0 and 1. The larger the value, the better the network device status. The pre-selected neural network is used for training and optimization.

[0034] A2. The device network score and activity value are combined to form network device feature information.

[0035] S103 : Obtain a target propensity score value by using a preset propensity determination algorithm according to the network device feature information and the initial propensity score value.

[0036] S104 , based on the target propensity scores of all target users, potential users are screened and their business interaction modes are differentiated, business interaction information corresponding to the business interaction mode is sent to the potential users, and feedback information from the potential users is obtained.

[0037] Specifically, target users whose target propensity score values ​​are greater than a preset propensity threshold are regarded as potential users.

[0038] S105: Input the potential user's feedback information into a pre-trained business type recognition model to obtain the potential user's preferred business type.

[0039] The pre-trained business type recognition model extracts keywords from user feedback information and determines the business type of potential users based on the keywords. For example, when a user feedbacks "I want to apply for a patent or trademark", the keyword "I want to apply" is captured to know that the user wants to handle related business. When the user feedbacks "I have some customer resources here, I want to cooperate with you", the keywords "resources, cooperation" are captured to know that the user wants to engage in distribution cooperation. When the user feedbacks "No problem" or "learning", the keyword "learning" or no keywords are captured to know that the user only wants to learn. The setting of keywords can be determined by big data of previous user consultations; therefore, the business type of potential users can be determined by the business type recognition model, which will not be elaborated in the present invention.

[0040] S106, repeat steps S101 to S105 until the preset cycle time is reached, count the preferred business types of all potential users to form an actual feedback business library, compare the target proportion of business types in the target business tweets with the actual proportion of business types in the actual feedback business library, and obtain the push value of each business type in the target business tweets.

[0041] Specifically, if the actual percentage of a business type in the actual feedback business database is greater than the target percentage for that business type, it means that the target business tweets have greater value for that business type. If the actual percentage of a business type in the actual feedback business database is less than the target percentage for that business type, it means that the target business tweets have less value for that business type. Therefore, based on the push value of the target business tweets for each business type, the target business tweets can be optimized, modified, and adjusted to maximize their economic value.

[0042] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: It not only considers the user's explicit interaction data, but also introduces implicit interaction data, such as browsing time and page scrolling depth. These data can reflect the user's interests and behavioral habits in more detail. By integrating network device feature information such as network status parameters, terminal hardware information and activity values, it can more accurately evaluate the user's interaction behavior in different devices and network environments, avoiding the limitations of a single data source.

[0043] By introducing diversified interactive behavior data and network device feature information, it can more comprehensively reflect the user's interest in the target business tweets. Accurate user propensity assessment provides strong support for the subsequent mining of potential users of the target business tweets, avoiding the waste of business interaction resources for the target business tweets, effectively improving the response rate of business interaction information, and also improving the accuracy and effectiveness of the value assessment of the target business tweets. By considering the influence of network devices and status, it can maintain a high evaluation accuracy in different devices and network environments, avoiding deviations in the propensity score value.

[0044] Example 2: Since different device networks and user activity may affect the calculation of the user's actual propensity for the target business tweet, resulting in a possible deviation between the initial propensity score and the actual value, the propensity determination algorithm is further limited.

[0045] Therefore, the embodiments of the present application are optimized based on the above embodiments.

[0046] In some embodiments, the preset trend determination algorithm includes: B1. Perform a weighted sum of the device network score and activity value in the network device feature information to obtain the interaction credibility value.

[0047] B2. If the interaction credibility value is lower than the preset threshold, execute step B3; if the interaction credibility value is higher than or equal to the preset threshold, use the initial propensity score value as the target propensity score value.

[0048] B3. Combining the initial propensity score value and the network device feature information to form a multivariate effect feature vector of the user's recommendation for the target service, and inputting it into a pre-trained propensity score correction model to obtain a correction guidance value.

[0049] B4. Use the correction guide value to correct the initial propensity score value to obtain the target propensity score value.

[0050] Specifically, the propensity score correction model is obtained as follows: C1. Collect the network device feature information and initial propensity score values ​​of different target users whose interaction credibility values ​​corresponding to different business tweets in history are lower than the preset threshold, and form multiple historical multivariate effect feature vectors.

[0051] C2. Label each historical multivariate effect feature vector, and set the label content as the correction guide value.

[0052] The correction guide value is obtained in the following way: D1. Based on each historical multivariate effect feature vector, obtain the immediate interest level of the corresponding user's feedback information.

[0053] Among them, the immediate interest level is determined based on the information effectiveness in the user's feedback information on the business tweet. The greater the information effectiveness, the greater the user's immediate interest level. If the user does not respond to the feedback information, the information effectiveness level is 0. The calculation method of the information effectiveness level can be determined based on the text correlation (similarity) between the feedback information and the business tweet, and the value range is between 0 and 1.

[0054] D2. Determine a correction guide value for the initial propensity score based on the initial propensity score value and its corresponding propensity threshold, the immediate interest level and its corresponding interest threshold.

[0055] The correction guidance value is an indicator used to adjust the initial propensity score value. It is determined based on the user's immediate interest and the comparison result of the initial propensity score value with the preset threshold. The purpose of the correction guidance value is to correct the deviation of the initial propensity score value that may be caused by network device feature information (such as device network score and activity value), so as to more accurately reflect the user's true interest in business tweets.

[0056] Among them, the propensity threshold and the interest threshold are both pre-set and determined according to actual conditions, and are used to define the propensity score value and the degree of immediate interest.

[0057] Specifically, step D2 includes: When the deviation direction between the initial propensity score and the corresponding propensity threshold is the same as the deviation direction between the immediate interest and the corresponding interest threshold, a correction guide value is obtained according to the first preset algorithm:

[0058] Among them, C is the correction guidance value, I is the immediate interest value, is the interest threshold, S is the initial propensity score value, is the tendency threshold, is a sign function that maintains the original deviation direction. k1 and k2 are pre-set weight values ​​used to adjust the influence of immediate interest and initial propensity score on the correction guidance value. k1 is greater than k2.

[0059] Therefore, when the immediate interest level and the initial propensity score are both higher or lower than their respective thresholds, the correction guidance value C is positive or negative, and its absolute value is large, which strengthens the original tendency or interest.

[0060] When the deviation direction between the initial propensity score and the corresponding propensity threshold is opposite to the deviation direction between the immediate interest and the corresponding interest threshold, the correction guidance value is obtained according to the second preset algorithm:

[0061] Among them, C is the correction guidance value, I is the immediate interest value, is the interest threshold, S is the initial propensity score value, is the tendency threshold, is a sign function that maintains the original deviation direction, and k3 and k4 are pre-set weight values. This part considers the relative deviation of the immediate interest degree from the interest threshold and reverses the deviation direction (because the directions are opposite) by multiplying the two sign functions. This part takes the smaller value of the two relative deviations as the harmonic term.

[0062] Therefore, when one of the immediate interest and initial propensity score values ​​is higher than the threshold and the other is lower than the threshold, the sign of the correction guidance value C is determined by the first part (negative or positive, depending on which deviation is larger and in the opposite direction), while the absolute value is limited by the harmonic term in the second part and will not be too large.

[0063] C3. Use the labeled historical multivariate effect feature vectors as the training set, select an appropriate machine learning algorithm (such as a neural network, support vector machine, etc.) as the basis of the propensity score correction model, use the training set for training and learning, optimize the model parameters, and obtain the propensity score correction model.

[0064] In summary, the device network score and activity value in the network device feature information are weighted and summed to obtain the interaction credibility value. Based on the comparison of the interaction credibility value with the preset threshold, a decision is made as to whether to use the initial propensity score directly as the target propensity score or to combine the initial propensity score with the network device feature information to form a multivariate effect feature vector. This is then input into the propensity score correction model to obtain a correction guide value, which is then used to correct the initial propensity score.

[0065] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: By introducing interactive credibility values ​​and correction guidance values, the initial propensity score values ​​are corrected in a more detailed and reasonable manner, taking into account the impact of different device networks and user activity on user propensity, making the evaluation results more accurate; through the propensity score correction model, the deviation of the initial propensity score value can be automatically adjusted, which improves the intelligence and automation level of the evaluation and can better adapt to changes in different users and scenarios; by considering the immediate interest level and the comparison results of the initial propensity score value with the preset threshold, the initial propensity score value can be corrected in a targeted manner to obtain a more accurate target propensity score value.

[0066] Example 3: Further limit the business interaction mode of potential users.

[0067] Therefore, the embodiments of the present application are optimized based on the above embodiments.

[0068] In step S104, the target business tweet is set with differentiated business interaction methods and business interaction information content for business interaction, and its corresponding interaction level is determined according to the target propensity score value. Each interaction level corresponds to a business interaction method, so as to differentiate the business interaction methods (emotion, form) of different potential users, improve the effectiveness of potential user feedback information, guide potential users to respond to more valuable questions or information, and facilitate subsequent identification and judgment of business types.

[0069] In some embodiments, step S104 specifically includes: S201, determining the interaction level: Based on a pre-set scoring interval, according to the target propensity score value of the potential user, matching the corresponding interaction level, each scoring interval corresponds to an interaction level.

[0070] Among them, the interaction level corresponds to the scoring interval one by one. For example, the scoring interval includes [ 、[ 、[ , corresponding to low level, intermediate level and high level respectively, a is smaller than b and is preset according to actual situation.

[0071] S202, matching business interaction methods: presetting one or more business interaction methods for each interaction level.

[0072] Among them, business interaction methods may include but are not limited to: interactive emotions and interactive depth.

[0073] For example, the business interaction method corresponding to the low level is set to a more enthusiastic question-and-answer format, and the business interaction method corresponding to the high level is set to a more in-depth business interaction. The business interaction methods are differentiated according to different levels to increase the participation of potential users and the effectiveness of feedback information.

[0074] For example: Low-level engagement: Set up a more enthusiastic, basic Q&A format, such as using a friendly greeting and providing a basic business introduction. Intermediate-level engagement: Provide personalized recommendations, combining historical user behavior and preferences to send exclusive offers. Advanced-level engagement: Set up more in-depth business interactions, such as customized interactive Q&A, invitations for in-depth communication, or provide exclusive VIP services.

[0075] Business interaction methods should vary based on the level of interaction to reflect differentiated services. Interaction emotion and interaction depth are two important dimensions of business interaction methods, but are not limited to these. These dimensions may also include interaction frequency and interaction format (e.g., text, voice, video), and are not limited to these in this invention.

[0076] S203, sending business interaction information: according to the interaction level of the potential user, select the corresponding business interaction method and send business interaction information to the potential user. The information content should match the interaction level to differentiate the business interaction experience of different potential users.

[0077] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: Target users whose target propensity score values ​​are greater than the preset propensity threshold are regarded as potential users, and their corresponding interaction levels are determined based on the target propensity score values. Each interaction level corresponds to a business interaction method, so as to differentiate the business interaction methods (emotion, form) of different potential users, improve the effectiveness of potential user feedback information, and guide potential users to respond to more valuable questions or information, which is convenient for subsequent identification and judgment of business types.

[0078] By determining the engagement level of potential users based on their target propensity scores and presetting corresponding business interaction methods for each engagement level, business interactions become more targeted and effective. Different engagement levels correspond to different business interaction methods, catering to the needs and preferences of different potential users, increasing their engagement and the effectiveness of their feedback. By differentiating the business interaction methods (emotion and form) for different potential users, we provide them with a more personalized service experience. This differentiated service approach can enhance potential users' interest in and recognition of business tweets, thereby increasing the response rate of business interaction messages.

[0079] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Various modifications and variations are readily apparent to those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A data processing method based on artificial intelligence, characterized in that: include: S101, collecting target user's interactive behavior data on target business tweets to obtain the target user's initial propensity score for the target business tweets; S102, obtaining network status parameters, terminal hardware information, and activity values ​​of target user interaction behavior data, and constructing network device feature information; S103, obtaining a target propensity score value using a preset propensity determination algorithm based on the network device feature information and the initial propensity score value; S104, based on the target propensity scores of all target users, screening potential users and differentiating the business interaction methods of the potential users, sending business interaction information corresponding to the business interaction methods to the potential users, and obtaining feedback information from the potential users; S105, inputting the potential user's feedback information into a pre-trained business type recognition model to obtain the potential user's preferred business type; S106, repeat steps S101 to S105 until the preset cycle time is reached, count the preferred business types of all potential users to form an actual feedback business library, compare the target proportion of business types in the target business tweets with the actual proportion of business types in the actual feedback business library, and obtain the push value of each business type in the target business tweets.

2. The artificial intelligence-based data processing method according to claim 1, wherein: The target users are all users who browse the target business tweets. The interactive behavior data includes the length of stay, page scroll depth and the duration of the like effect. The initial propensity score value is calculated according to the following formula: Among them, S is the initial propensity score value, T is the length of stay, and D is the page scroll depth. is the maximum page scroll depth, The duration of the like effect. 、 and are the weight factors of the initial propensity score value, namely, the length of stay, the page scroll depth and the time of the like.

3. The artificial intelligence-based data processing method according to claim 1, wherein: The constructing of network device feature information specifically includes: A1. Obtain network status parameters and terminal hardware information, and use the pre-trained device network scoring model to obtain the device network score. A2. The device network score and activity value are combined to form network device feature information.

4. The artificial intelligence-based data processing method according to claim 3, wherein: The preset tendency determination algorithm includes: B1. Perform a weighted sum of the device network score and activity value in the network device feature information to obtain an interaction credibility value; B2. If the interaction credibility value is lower than the preset threshold, execute step B3; if the interaction credibility value is higher than or equal to the preset threshold, use the initial propensity score value as the target propensity score value; B3. Combining the initial propensity score value and network device feature information to form a multivariate effect feature vector for the user's recommendation of the target service, and inputting it into a pre-trained propensity score correction model to obtain a correction guidance value; B4. Use the correction guide value to correct the initial propensity score value to obtain the target propensity score value.

5. The artificial intelligence-based data processing method according to claim 4, characterized in that: The propensity score correction model is obtained as follows: C1. Collect network device feature information and initial propensity score values ​​of different target users whose interaction credibility values ​​for different business tweets are lower than the preset threshold in history, and form multiple historical multivariate effect feature vectors. C2. Label each historical multivariate effect feature vector, and set the label content as the correction guide value; C3. Use the labeled historical multivariate effect feature vectors as a training set, use the training set to train and learn the pre-selected neural network structure, optimize the model parameters, and obtain the propensity score correction model.

6. The artificial intelligence-based data processing method according to claim 5, wherein: The correction guide value is obtained in the following manner: D1. Based on each historical multivariate effect feature vector, obtain the immediate interest level of the corresponding user's feedback information; D2. Determine a correction guide value for the initial propensity score based on the initial propensity score value and its corresponding propensity threshold, the immediate interest level and its corresponding interest threshold.

7. The artificial intelligence-based data processing method according to claim 6, wherein: Step D2 specifically includes: When the deviation direction between the initial propensity score value and the corresponding propensity threshold value is the same as the deviation direction between the immediate interest value and the corresponding interest threshold value, a correction guide value is obtained according to the first preset algorithm: Among them, C is the correction guidance value, I is the immediate interest value, is the interest threshold, S is the initial propensity score value, is the tendency threshold, is a sign function that maintains the original deviation direction. k1 and k2 are pre-set weight values ​​used to adjust the influence of immediate interest and initial propensity score on the correction guidance value. k1 is greater than k2.

8. The artificial intelligence-based data processing method according to claim 6, wherein: Said D2 further comprises: When the deviation direction between the initial propensity score and the corresponding propensity threshold is opposite to the deviation direction between the immediate interest and the corresponding interest threshold, the correction guidance value is obtained according to the second preset algorithm: Among them, C is the correction guidance value, I is the immediate interest value, is the interest threshold, S is the initial propensity score value, is the tendency threshold, is a sign function that maintains the original deviation direction, and k3 and k4 are pre-set weight values.

9. The artificial intelligence-based data processing method according to claim 7, wherein: The S104 specifically includes: S201, matching corresponding interaction levels based on the target propensity scores of potential users based on pre-set score intervals, where each score interval corresponds to an interaction level; S202, presetting one or more business interaction modes for each interaction level; S203: Select a corresponding business interaction method according to the interaction level of the potential user, and send business interaction information to the potential user.

10. The artificial intelligence-based data processing method according to claim 9, wherein: The interaction level corresponds to the scoring interval one by one, and the scoring interval includes [ 、[ and[ , corresponding to low level, medium level and high level respectively, a is smaller than b and is preset according to actual situation.

Images