Data processing method and device based on user new media APP interactive operation

Abstract

The invention discloses a data processing method and device based on user new media APP interaction operation, and the method comprises the steps: responding to an interaction instruction of a user for a comment box of a target object, and inputting the content into the comment box in real time; the target object is a work browsed by the user in the new media APP, and the work is used for describing the target product; identifying whether semantic clusters related to the content of the real-time input comment box exist in semantic clusters corresponding to all themes in a preset dynamic theme library or not; if yes, obtaining a target theme to which a semantic cluster related to the content input into the comment box in real time belongs, intercepting a regional screenshot of the comment box, and extracting all user comment texts; according to a preset quantization parameter and the user comment text, determining a demand label of the user for the target product; and storing a ternary mapping relationship among the product identifier of the target product, the target theme and the demand label. Therefore, by adopting the method and the device, high accuracy of the acquired data can be ensured, and real demands and behavior modes of the user can be comprehensively reflected.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a data processing method and device based on user new media APP interactive operations. Background Art

[0002] In today's digital age, new media apps (such as social media and e-commerce) have become important platforms for people to obtain information, express opinions and interact. User interactions in apps, such as comments, likes, and sharing, contain rich information, which is extremely valuable for product optimization, market research, and personalized recommendations. Therefore, how to efficiently and accurately collect and process user interaction data in new media apps has become the focus of many companies and research institutions.

[0003] In related technologies, data collection technology mainly relies on users to actively submit information or obtain data through a single sensor. For example, some apps collect user feedback through questionnaires, and users need to manually fill in and submit the questionnaires. Questionnaires and other methods that rely on users to actively submit information are obviously passive, with low user participation, limited coverage, and are easily affected by user subjective factors, resulting in deviations in the collected data, making it difficult to fully reflect the real needs and behavior patterns of users. Summary of the invention

[0004] The embodiments of the present application provide a data processing method and device based on user new media APP interactive operations. In order to have a basic understanding of some aspects of the disclosed embodiments, a simple summary is given below. This summary is not a general review, nor is it intended to identify key / important components or describe the scope of protection of these embodiments. Its only purpose is to present some concepts in a simple form as a preface to the detailed description that follows.

[0005] In a first aspect, an embodiment of the present application provides a data processing method based on a user's new media APP interactive operation, the method comprising: In response to the user's interactive instruction on the comment box of the target object, the content input by the user on the comment box is obtained in real time; wherein the target object is the work browsed by the user in the new media APP, and the work is used to describe the target product; Performing contextual semantic analysis on the content of the real-time input comment box to identify whether there is a semantic cluster related to the content of the real-time input comment box in the semantic clusters corresponding to each theme in the preset dynamic theme library; wherein the preset dynamic theme library is regularly updated based on multiple original comments existing in the work within a preset period; In the case where there is a semantic cluster related to the content of the real-time input comment box, obtaining a target topic to which the semantic cluster related to the content of the real-time input comment box belongs, taking a regional screenshot of the comment box, and extracting all user comment texts in the regional screenshot; Determine the user's demand label for the target product based on the preset quantitative parameters and user comment text; In the database, a ternary mapping relationship between the product identification of the target product, the target subject and the demand label is stored.

[0006] In a second aspect, an embodiment of the present application provides a data processing device based on user new media APP interactive operation, the device comprising: A response acquisition module, used to respond to the user's interactive instruction on the comment box of the target object, and acquire the content input by the user on the comment box in real time; wherein the target object is the work browsed by the user in the new media APP, and the work is used to describe the target product; The semantic cluster determination module is used to perform contextual semantic analysis on the content of the real-time input comment box to identify whether there is a semantic cluster related to the content of the real-time input comment box in the semantic clusters corresponding to each theme in the preset dynamic theme library; wherein the preset dynamic theme library is regularly updated based on multiple original comments existing in the work within a preset period; A text extraction module is used to obtain a target topic to which the semantic cluster related to the content of the real-time input comment box belongs, if there is a semantic cluster related to the content of the real-time input comment box, and to capture a regional screenshot of the comment box, and to extract all user comment texts in the regional screenshot; A demand label determination module is used to determine the user's demand label for the target product based on preset quantitative parameters and user comment text; The ternary mapping relationship storage module is used to store the ternary mapping relationship between the product identification of the target product, the target subject and the demand label in the database.

[0007] The technical solution provided by the embodiments of the present application may have the following beneficial effects: In the embodiment of the present application, on the one hand, by acquiring user input content in real time and combining it with the contextual semantic analysis of each topic in the preset dynamic theme library, the system can identify semantic clusters in user comments in real time, not just keywords. This semantic-based analysis method can effectively reduce misjudgments caused by the diversity of language expressions, thereby ensuring the accuracy of the collected data. At the same time, the system regularly updates semantic clusters through the preset dynamic theme library, which can adapt to changes in user focus and further improve the accuracy and timeliness of semantic recognition. On the other hand, the system determines the demand label based on preset quantitative parameters and user comment text, and stores the ternary mapping relationship between the product identification of the target product, the target theme and the demand label. This structured data storage method is not only convenient for subsequent analysis, but also can fully reflect the user's real needs and behavior patterns through multi-dimensional data fusion.

[0008] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0010] Figure 1 This is a method flow diagram of a data processing method based on user new media APP interactive operation provided by an embodiment of the present application; Figure 2 This is a schematic diagram of a comment box page provided in an embodiment of the present application; Figure 3 This is a schematic diagram of prompt word information provided by an embodiment of the present application; Figure 4 This is a schematic diagram of a demand tag visualization interface provided by an embodiment of the present application; Figure 5 This is another schematic diagram of a demand tag visualization interface provided in an embodiment of the present application; Figure 6 It is a flowchart of a model training method for a semantic cluster analysis model provided in an embodiment of the present application; Figure 7 It is a structural schematic diagram of a data processing device based on user new media APP interactive operation provided in an embodiment of the present application; Figure 8 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] The following description and the accompanying drawings sufficiently illustrate specific embodiments of the present application to enable those skilled in the art to practice them.

[0012] It should be clear that the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0013] When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Instead, they are only examples of devices and methods consistent with some aspects of the present application as detailed in the attached claims.

[0014] In the description of the present application, it should be understood that the terms "first", "second", etc. are used for descriptive purposes only and should not be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances. In addition, in the description of the present application, unless otherwise specified, "multiple" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship.

[0015] At present, data collection technology mainly relies on users to actively submit information or obtain data through a single sensor. For example, some apps collect user feedback through questionnaires, and users need to manually fill in the questionnaire and submit it.

[0016] The inventors realized that methods such as questionnaires that rely on users' active submission are obviously passive, have low user participation, limited coverage, and are easily affected by user subjective factors, resulting in bias in the collected data and making it difficult to fully reflect users' real needs and behavior patterns.

[0017] In order to solve the above problems, the present application provides a data processing method and device based on user new media APP interactive operation to solve the problems existing in the above-mentioned related technical problems. In the embodiment of the present application, on the one hand, by obtaining user input content in real time and combining the context semantic analysis of each topic in the preset dynamic theme library, the system can identify the semantic clusters in user comments in real time, not just keywords. This semantic-based analysis method can effectively reduce the misjudgment caused by the diversity of language expression, thereby ensuring the accuracy of the collected data. At the same time, the system regularly updates the semantic clusters through the preset dynamic theme library, which can adapt to the changes in user focus and further improve the accuracy and timeliness of semantic recognition. On the other hand, the system determines the demand label according to the preset quantitative parameters and the user comment text, and stores the ternary mapping relationship between the product identification of the target product, the target theme and the demand label. This structured data storage method is not only convenient for subsequent analysis, but also can be through multi-dimensional data fusion, so as to fully reflect the real needs and behavior patterns of users. The following exemplary embodiments are used for detailed description.

[0018] The following will be combined with the attached Figure 1 -Attached Figure 6 , a data processing method based on user new media APP interactive operation provided by an embodiment of the present application is introduced in detail. The method can be implemented by relying on a computer program and can be run on a data processing device based on user new media APP interactive operation based on the von Neumann system. The computer program can be integrated into an application or run as an independent tool application.

[0019] See also Figure 1 , which is a flow chart of a data processing method based on user new media APP interactive operation according to an embodiment of the present application. Figure 1 As shown, the method of the embodiment of the present application may include the following steps: S101, in response to an interactive instruction of a user on a comment box of a target object, obtaining content input by the user on the comment box in real time; wherein the target object is a work browsed by the user in a new media APP, and the work is used to describe a target product; Among them, users refer to individuals who browse content and interact in new media apps. The target object is the specific content that users browse in new media apps, which is a work. The comment box is an input box in the app interface for users to enter comments. Interaction instructions are operations performed by users in the comment box, such as clicking, entering text, etc. Content is the specific text information entered by users in the comment box. New media apps are applications such as social media and e-commerce, where users browse and interact through these platforms. Works are the specific content of the target object, which may be product introductions, articles, videos, etc. The target product is the product described in the work.

[0020] In some embodiments of the present application, when a user browses a work describing a target product in a new media APP, if the user triggers the comment box to input a comment, the system will respond to the user's operation (interaction instruction) in real time and obtain the content input by the user for the comment box.

[0021] For example, user Xiaoming is browsing the introduction page of a smart watch in a certain e-commerce APP. There are product pictures, function introductions, and a comment box for users to post comments on the page. Xiaoming triggers the comment box, and a comment box page pops up. For example Figure 2 as shown, Xiaoming enters in the comment box: "The appearance design of this smart watch is very fashionable, but the battery life is too short", and the system will immediately respond to Xiaoming's operation and obtain the comment content he entered in real time for subsequent analysis.

[0022] S102. Perform context semantic analysis on the content input into the comment box in real time to identify whether there is a semantic cluster related to the content input into the comment box in real time in the semantic clusters corresponding to each theme in the preset dynamic theme library; wherein, the preset dynamic theme library is obtained by being regularly updated according to multiple original comments existing in the works within a preset period; Among them, context semantic analysis is used to, when each character is input into the comment box, combine the character with the characters input historically for semantic analysis. Since the processing speed of the algorithm is much faster than the speed at which the user inputs characters, every time the user inputs a character, it can be combined with the characters input historically to form new content for real-time analysis, thereby improving the real-time performance of the analysis. The preset dynamic theme library is a database that is regularly updated according to multiple original comments collected within a preset period, and is used to store and manage the themes related to the product and the semantic clusters corresponding to the themes. A semantic cluster is a group of related words and phrases around a certain theme or concept, and these words are semantically similar or relevant. The original comments are the historical views expressed by different users in the new media APP for a certain product or content. The preset period is the time interval for regularly updating the dynamic theme library.

[0023] In some embodiments of the present application, after the client obtains the content input by the user for the comment box, for example, at the current moment, the content input by Xiaoming is the last character "le" of the sentence "The appearance design of this smart watch is very fashionable, but the battery life is too short". At this time, the client can splice the last character "le" with the previously input historically "The appearance design of this smart watch is very fashionable, but the battery life is too short" to obtain the content input into the comment box in real time, that is, "The appearance design of this smart watch is very fashionable, but the battery life is too short".

[0024] For example, the system performs contextual semantic analysis on the comment input by Xiao Ming, and searches the preset dynamic theme library for semantic clusters related to the sentence "This smart watch has a fashionable design, but the battery life is too short". Assuming that each theme in the dynamic theme library corresponds to the following semantic clusters (design: fashionable, design sense, high value, fashionable appearance; battery life: short battery life, durable battery), the system can identify that the sentence "This smart watch has a fashionable design, but the battery life is too short" exists in two semantic clusters in the dynamic theme library.

[0025] In an embodiment of the present application, a specific process of identifying whether there is a semantic cluster related to the content of the real-time input comment box in the semantic clusters corresponding to each topic in the preset dynamic topic library includes: traversing and obtaining the semantic clusters corresponding to each topic from the preset dynamic topic library; extracting the semantic feature vector of the content of the real-time input comment box as the center point vector; using the embedding vector of the semantic cluster corresponding to each topic as multiple edge vectors of the center point vector, and establishing a relationship between the center point vector and each edge vector to obtain multiple objects to be analyzed; for each object to be analyzed, calculating the semantic similarity between the center point vector and each edge vector to obtain multiple semantic similarities of each object to be analyzed; calculating the similarity mean of the multiple semantic similarities of each object to be analyzed to obtain the target semantic similarity of each object to be analyzed; when there is an object to be analyzed whose target semantic similarity is greater than a preset threshold, determining that there is a semantic cluster related to the content of the real-time input comment box in the semantic clusters corresponding to each topic in the preset dynamic topic library; or, when there is no object to be analyzed whose target semantic similarity is greater than the preset threshold, determining that there is no semantic cluster related to the content of the real-time input comment box in the semantic clusters corresponding to each topic in the preset dynamic topic library.

[0026] For example, the dynamic topic library contains the following topics and corresponding semantic clusters: Topic: Product Appearance, semantic cluster embedding vector: [0.8, 0.1, 0.2] (embedding vectors for words such as "fashionable", "sense of design", and "good-looking"). Topic: Product Performance, semantic cluster embedding vector: [0.1, 0.7, 0.3] (embedding vectors for words such as "short battery life" and "durable battery"). Topic: Price, semantic cluster embedding vector: [0.2, 0.2, 0.6] (embedding vectors for words such as "cost-effective" and "reasonable price"). Xiao Ming's comment is: "The appearance design of this smart watch is very fashionable, but the battery life is too short." The system extracts the semantic feature vector of this comment through natural language processing (NLP) technology. Assume that the center point vector obtained is: [0.7, 0.4, 0.5]. The semantic cluster embedding vectors corresponding to each topic in the dynamic topic library are used as edge vectors, and a relationship is established with the center point vector to obtain multiple objects to be analyzed: Object 1 to be analyzed: Center point vector: [0.7, 0.4, 0.5], Edge vector (product appearance): [0.8, 0.1, 0.2]. Object 2 to be analyzed: Center point vector: [0.7, 0.4, 0.5], Edge vector (product performance): [0.1, 0.7, 0.3]. Object 3 to be analyzed: Center point vector: [0.7, 0.4, 0.5], Edge vector (price): [0.2, 0.2, 0.6].

[0027] At this time, for each object to be analyzed, the semantic similarity between the center point vector and the edge vector is calculated (for example, using cosine similarity). The calculation results are as follows: Object 1 to be analyzed (product appearance): similarity = 0.85, Object 2 to be analyzed (product performance): similarity = 0.78, Object 3 to be analyzed (price): similarity = 0.35. Assume that the preset threshold is 0.6. According to the target semantic similarity: Product appearance (0.85)>0.6, there is a related semantic cluster. Product performance (0.78)>0.6, there is a related semantic cluster. Price (0.35)<0.6, there is no related semantic cluster.

[0028] At this point, the system can determine that Xiao Ming's comments are related to the semantic clusters in the "product appearance" topic (such as "fashionable"). Xiao Ming's comments are related to the semantic clusters in the "product performance" topic (such as "short battery life"). Xiao Ming's comments are not related to the semantic clusters in the "price" topic.

[0029] In the embodiment of the present application, the preset dynamic theme library can be continuously updated, and the specific process of the update is: obtain multiple original comments existing in the work within the preset period; analyze the original comments that are not related to the themes in the preset dynamic theme library from the multiple original comments as the comments to be analyzed; input each comment to be analyzed into the pre-trained semantic cluster analysis model, and output the semantic cluster corresponding to each comment to be analyzed; wherein the pre-trained semantic cluster analysis model is a data model obtained by machine learning based on the historical user comment data stored in the preset time period; subject-classify the semantic cluster corresponding to each comment to be analyzed to obtain the theme corresponding to each comment to be analyzed; from the themes corresponding to each comment to be analyzed, count the number of themes of the same theme; when the number of themes of the same theme is greater than or equal to the preset number threshold, store the binary mapping relationship between the same theme and all the semantic clusters corresponding to the same theme in the preset dynamic theme library. wherein, by regularly updating the semantic clusters through the preset dynamic theme library, it is possible to adapt to the changes in user focus and further improve the accuracy and timeliness of semantic recognition.

[0030] For example, the system obtains user reviews for a smart watch in the past week, and the reviews are as follows: 1. "This smartwatch has a stylish design"; 2. "The battery life is too short, I hope it can be improved"; 3. "The price is a bit expensive and the cost performance is not high"; 4. "The watch is very waterproof"; 5. "The operation interface is complicated and not very user-friendly."

[0031] Assume that the existing topics in the dynamic topic library are "product appearance" and "product performance". The system analyzes these comments and finds that the third and fifth comments are not related to the existing topics, so they are marked as comments to be analyzed. The comments to be analyzed are input into the semantic cluster analysis model, and the model outputs the following: The semantic clusters corresponding to the third comment are: price, cost-effectiveness; the semantic clusters corresponding to the fifth comment are: operating interface, user experience.

[0032] The system classifies these semantic clusters into corresponding topics: the third comment: the topic is "price", the fifth comment: the topic is "user experience".

[0033] Count the number of comments for each topic: topic "price": 1 comment, topic "user experience": 1 comment. Assume that the preset number threshold is 2, and the number of comments for no topic currently reaches the threshold. Assume that after a period of time, the system continues to collect comments and finds that the number of comments for the topic "price" reaches 3 (exceeding the threshold), then the topic "price" and its corresponding semantic cluster are stored in the dynamic topic library. Among them, the updated binary mapping relationship table of the dynamic topic library is shown in Table 1.

[0034] Table 1

[0035] In an embodiment of the present application, the specific process of generating a pre-trained semantic cluster analysis model includes: collecting historical user comment data stored within a preset time period from a new media APP, the historical user comment data covering multiple works; performing data preprocessing on the historical user comment data to obtain a vocabulary sequence corresponding to each historical user comment data, the data preprocessing including text cleaning, word segmentation, stop word removal and part-of-speech tagging; capturing semantic information of the vocabulary sequence corresponding to each historical user comment data to convert the vocabulary sequence corresponding to each historical user comment data into a high-dimensional vector to obtain multiple vocabulary features of each sample comment data; constructing a semantic cluster of each sample comment data based on the multiple vocabulary features of each sample comment data; using the semantic cluster of each sample comment data as a label, annotating the vocabulary sequence corresponding to each historical user comment data to obtain a model training sample; creating a semantic cluster analysis model; inputting the model training sample into the semantic cluster analysis model for machine learning, and outputting a model loss value; when the model loss value reaches the minimum, generating a pre-trained semantic cluster analysis model.

[0036] Among them, text cleaning includes removing noises in the comments, such as HTML tags, special characters, emojis, etc. Word segmentation includes splitting the comment text into separate sequences of words. For example, splitting "This mobile phone is very useful" into "This / mobile phone / very / useful". Stop word removal includes removing common words that contribute less to semantics, such as "de", "shi", "he", etc. Part-of-speech tagging includes tagging the part of speech (noun, verb, adjective, etc.) for each word. For example, the original comment: "The battery life of this mobile phone is really good!", after preprocessing: ["This", "mobile phone", "battery", "life", "really", "good"].

[0037] Among them, semantic information capture converts the sequence of words into high-dimensional vectors through algorithms (such as Word2Vec, BERT, etc.), and these vectors can capture the semantic information of the words. High-dimensional vectors mean that each word or comment is represented as a point in a high-dimensional space, and similar words or comments are closer in the space. For example, the sequence of words: ["mobile phone", "battery", "life"], the converted high-dimensional vectors are [[0.1, 0.2, 0.3], [0.4, 0.5, 0.6], [0.7, 0.8, 0.9]].

[0038] Among them, the semantic cluster of each comment is used as a label for supervised learning. The labeled data is used to train the semantic cluster analysis model. For example, the comment is ["mobile phone", "battery", "life"], and the semantic cluster label is ["battery", "life"], and the final training sample of this comment is ([0.1, 0.2, 0.3], ["battery", "life"]).

[0039] Among them, during the training process, the model loss value gradually decreases from 1.0 to 0.1. At this time, the model training is completed, and a semantic cluster analysis model is generated.

[0040] In an embodiment of the present application, the specific process of constructing a semantic cluster of each sample comment data according to multiple lexical features of each sample comment data is as follows: initialize multiple lexical features of each sample comment data to mark multiple lexical features as unvisited; traverse the first unvisited lexical feature among the multiple lexical features of each sample comment data, and use the traversed lexical feature as the lexical feature to be analyzed, and mark the state of the lexical feature to be analyzed as visited; query all unvisited first lexical features in the field of the lexical feature to be analyzed; calculate the similarity between the lexical feature to be analyzed and each first lexical feature; and mark the first lexical feature whose similarity is greater than or equal to a preset similarity threshold as the first lexical feature to be analyzed. The lexical features and the lexical features to be analyzed are grouped into one cluster to obtain a cluster for each sample comment data; the step of traversing the first unvisited lexical feature among multiple lexical features of each sample comment data is continued until multiple lexical features are all marked as visited to obtain multiple clusters for each sample comment data; the lexical features in each cluster of each sample comment data are reversely converted to obtain a lexical sequence for each cluster of each sample comment data; the lexical sequence for each cluster of each sample comment data is input into a preset large language model to analyze the representative vocabulary of each cluster of each sample comment data to obtain a semantic cluster for each sample comment data.

[0041] The calculation formula of similarity is: ; in, is the similarity, For the The first lexical feature, is the vocabulary feature to be analyzed, is the number of features of the first vocabulary feature.

[0042] Among them, the model training samples include the semantic clusters of each sample comment data and each sample comment data; the semantic cluster analysis model includes an acquisition module, a probability fitting module, a probability prediction module, a cross entropy loss value calculation module, a model loss value calculation module and a loss value output module.

[0043] In an embodiment of the present application, the model training samples are input into the semantic cluster analysis model for machine learning, and the specific process of outputting the model loss value is: the acquisition module obtains the semantic cluster of each sample comment data; the probability fitting module fits the first probability distribution of the semantic cluster of each sample comment data; the probability prediction module predicts the second probability distribution of each sample comment data; the cross-entropy loss value calculation module calculates the cross-entropy loss value of each sample comment data according to the first probability distribution and the second distribution probability; the model loss value calculation module calculates the cross-entropy loss value of the model training sample according to the cross-entropy loss value of each sample comment data as the model loss value; the loss value output module outputs the model loss value.

[0044] Among them, the loss function of the model loss value is:

[0045] in, is the model loss value, is the number of sample comment data in the model training sample, is the number of semantic clusters, It is The sample The true label of a semantic cluster. The samples belong to semantic clusters, then =1, otherwise =0, The model predicts The samples belong to The probability of a semantic cluster.

[0046] S103, if there is a semantic cluster related to the content of the real-time input comment box, obtain the target topic to which the semantic cluster related to the content of the real-time input comment box belongs, take a regional screenshot of the comment box, and extract all user comment texts in the regional screenshot; In an embodiment of the present application, the system first identifies whether the comment content input by the user is related to a preset semantic cluster through semantic analysis. If there is a correlation, determine the topic to which the semantic cluster related to the user input content belongs. For example, if the user comment is related to the semantic cluster of "battery life", the comment belongs to the topic of "product performance". The system captures the screen area where the comment box is located and generates a screenshot. The purpose of this step is to capture the visual information of the comment box and its related products. The system extracts the text content of all user comments from the screenshot through OCR (optical character recognition) technology. This step can obtain the comments entered by the user and further responses to the comments made by other users. Relevant comments from other users can assist in comprehensive analysis.

[0047] For example, the system performs semantic analysis on Xiao Ming's comments and identifies the following semantic clusters: "Appearance design" (related to the "Product appearance" theme), "Battery life" (related to the "Product performance" theme). The system determines the topics related to Xiao Ming's comments: For "Appearance design", the target topic is "Product appearance". For "Battery life", the target topic is "Product performance". The system automatically captures the screen area where the comment box is located and generates a screenshot. This screenshot may contain the comment box, Xiao Ming's comment content, and comments from other users. The system extracts the text content of all user comments from the screenshot through OCR technology. Suppose the screenshot contains the following comments: Xiao Ming's comment: "The appearance design of this smart watch is very fashionable, but the battery life is too short", comments from other users: "Indeed, the battery life is a problem", "The appearance design is great".

[0048] S104, determining the user's demand label for the target product according to the preset quantitative parameters and the user's comment text; The preset quantitative parameters include an emotion threshold and an attention threshold.

[0049] In an embodiment of the present application, the specific process of determining the user's demand label for the target product based on preset quantitative parameters and user comment text includes: performing sentiment tendency analysis on the user comment text through a sentiment analysis tool to obtain the user's sentiment polarity score; determining the user's sentiment tendency descriptor based on the sentiment threshold and the user's sentiment polarity score; matching the user comment text with a preset keyword template through a TF-IDF algorithm to determine multiple keywords in the user comment text; counting the length of the user comment text as the user's attention to the target product; determining the user's attention tendency descriptor based on the attention threshold and the attention; constructing a prompt word for the target product based on the sentiment tendency descriptor, the attention tendency descriptor and multiple keywords, the prompt word is used to guide the preset large language model to generate the user's demand label for the target product; inputting the prompt word into the preset large language model to output the user's demand label for the target product; wherein the preset large language model is ChatGPT.

[0050] For example, use a sentiment analysis tool to analyze Xiao Ming's comment and the comments of other users who responded to the comment to obtain the sentiment polarity score. Assume that the scores output by the sentiment analysis tool are: positive sentiment: 0.7, negative sentiment: 0.6, neutral sentiment: 0.1. The sentiment polarity score indicates that the comment contains both positive sentiment ("the appearance design is very fashionable") and negative sentiment ("the battery life is too short"). According to the preset sentiment threshold (assuming that the positive sentiment threshold is 0.8 and the negative sentiment threshold is 0.6), determine the sentiment tendency descriptor. The negative sentiment score of 0.6 reaches the threshold, so the sentiment tendency descriptor is "negative". Use the TF-IDF algorithm to extract keywords in the comment. Assuming that the preset keyword template includes "appearance design", "battery life", "price", etc., the extraction result is: Keywords: ["appearance design", "battery life"]. Count the length of Xiao Ming's comment (in characters or words). Assume that the comment length is 35 characters. According to the preset attention threshold (assuming that the high attention threshold is 30 characters), the attention tendency description word is determined. The comment length of 35 characters exceeds the threshold, so the attention tendency description word is "high attention". At this time, the prompt word of the target product is constructed to guide the large language model to generate demand labels. Prompt words such as Figure 3 The above information is input into the preset large language model (such as ChatGPT), and the model outputs the user's demand label for the target product as follows: "Users' demand labels for smart watches: satisfied with the appearance design, but hope to improve battery life."

[0051] S105, storing the ternary mapping relationship between the product identification of the target product, the target theme and the demand tag in the database.

[0052] Among them, the product identification of the target product is that each product has a unique identifier (such as product ID) in the database, which is used to distinguish different products. The target topic is the topic involved in the user review, such as "product appearance", "product performance", "price", etc. The demand label is a summary description of the user's needs, such as "satisfactory appearance design", "battery life needs to be optimized", etc. The ternary mapping relationship is to store the relationship between the product identification, topic and demand label in the form of a triple, which is convenient for quick query and analysis.

[0053] The ternary mapping relationship is shown in Table 2.

[0054] Table 2

[0055] In some embodiments of the present application, after storing the ternary mapping relationship between the product identification of the target product, the target subject, and the demand tag, in some visualization display scenarios, a data query request inputted by the client is received, and the data query request carries the product identification to be analyzed and the subject to be analyzed; all demand tags that satisfy the product identification to be analyzed and the subject to be analyzed are obtained from the ternary mapping relationship; all identical demand tags are aggregated to obtain non-repeated demand tags; and non-repeated demand tags are imported into the visualization component to visualize the non-repeated demand tags. Among them, the product identification to be analyzed and the subject to be analyzed are selected and submitted in the demand tag visualization interface, and the demand tag visualization interface is, for example, Figure 4 As shown, the display result of the visual display of the demand label is as follows: Figure 5 shown.

[0056] In the embodiment of the present application, on the one hand, by acquiring user input content in real time and combining it with the contextual semantic analysis of each topic in the preset dynamic theme library, the system can identify semantic clusters in user comments in real time, not just keywords. This semantic-based analysis method can effectively reduce misjudgments caused by the diversity of language expressions, thereby ensuring the accuracy of the collected data. At the same time, the system regularly updates semantic clusters through the preset dynamic theme library, which can adapt to changes in user focus and further improve the accuracy and timeliness of semantic recognition. On the other hand, the system determines the demand label based on preset quantitative parameters and user comment text, and stores the ternary mapping relationship between the product identification of the target product, the target theme and the demand label. This structured data storage method is not only convenient for subsequent analysis, but also can fully reflect the user's real needs and behavior patterns through multi-dimensional data fusion.

[0057] See also Figure 6 , which is a flow chart of a model training method for a semantic cluster analysis model provided in an embodiment of the present application. Figure 6 As shown, the method of the embodiment of the present application may include the following steps: S201, collecting historical user comment data stored in a preset time period from the new media APP, where the historical user comment data covers multiple works; S202, performing data preprocessing on the historical user comment data to obtain a vocabulary sequence corresponding to each piece of historical user comment data, wherein the data preprocessing includes text cleaning, word segmentation, stop word removal, and part-of-speech tagging; S203, capturing semantic information of the vocabulary sequence corresponding to each piece of historical user comment data, so as to convert the vocabulary sequence corresponding to each piece of historical user comment data into a high-dimensional vector, and obtain multiple vocabulary features of each sample comment data; S204, constructing a semantic cluster of each sample comment data according to multiple lexical features of each sample comment data; S205, using the semantic cluster of each sample comment data as a label, annotating the vocabulary sequence corresponding to each historical user comment data, and obtaining a model training sample; S206, creating a semantic cluster analysis model; S207, inputting the model training sample into the semantic cluster analysis model for machine learning, and outputting the model loss value; S208, when the model loss value reaches the minimum, a pre-trained semantic cluster analysis model is generated.

[0058] In the embodiment of the present application, on the one hand, by acquiring user input content in real time and combining it with the contextual semantic analysis of each topic in the preset dynamic theme library, the system can identify semantic clusters in user comments in real time, not just keywords. This semantic-based analysis method can effectively reduce misjudgments caused by the diversity of language expressions, thereby ensuring the accuracy of the collected data. At the same time, the system regularly updates semantic clusters through the preset dynamic theme library, which can adapt to changes in user focus and further improve the accuracy and timeliness of semantic recognition. On the other hand, the system determines the demand label based on preset quantitative parameters and user comment text, and stores the ternary mapping relationship between the product identification of the target product, the target theme and the demand label. This structured data storage method is not only convenient for subsequent analysis, but also can fully reflect the user's real needs and behavior patterns through multi-dimensional data fusion.

[0059] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0060] See also Figure 7 , which shows a schematic diagram of the structure of a data processing device based on user new media APP interactive operation provided by an exemplary embodiment of the present application. The data processing device based on user new media APP interactive operation can be implemented as all or part of an electronic device through software, hardware or a combination of both. The device 1 includes a response acquisition module 10, a semantic cluster determination module 20, a text extraction module 30, a demand label determination module 40, and a ternary mapping relationship storage module 50.

[0061] The response acquisition module 10 is used to respond to the user's interactive instruction on the comment box of the target object and acquire the content input by the user on the comment box in real time; wherein the target object is the work browsed by the user in the new media APP, and the work is used to describe the target product; The semantic cluster determination module 20 is used to perform contextual semantic analysis on the content of the real-time input comment box to identify whether there is a semantic cluster related to the content of the real-time input comment box in the semantic clusters corresponding to each theme in the preset dynamic theme library; wherein the preset dynamic theme library is regularly updated based on multiple original comments existing in the work within a preset period; The text extraction module 30 is used to obtain the target topic to which the semantic cluster related to the content of the real-time input comment box belongs, and to capture a regional screenshot of the comment box, and to extract all user comment texts in the regional screenshot, if there is a semantic cluster related to the content of the real-time input comment box. The demand label determination module 40 is used to determine the user's demand label for the target product according to the preset quantitative parameters and the user's comment text; The ternary mapping relationship storage module 50 is used to store the ternary mapping relationship between the product identification of the target product, the target theme and the requirement tag in the database.

[0062] It should be noted that the data processing device based on user new media APP interactive operation provided in the above embodiment only uses the division of the above-mentioned functional modules as an example when executing the data processing method based on user new media APP interactive operation. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the data processing device based on user new media APP interactive operation provided in the above embodiment and the data processing method based on user new media APP interactive operation belong to the same concept. The implementation process is detailed in the method embodiment, which will not be repeated here.

[0063] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0064] In the embodiment of the present application, on the one hand, by acquiring user input content in real time and combining it with the contextual semantic analysis of each topic in the preset dynamic theme library, the system can identify semantic clusters in user comments in real time, not just keywords. This semantic-based analysis method can effectively reduce misjudgments caused by the diversity of language expressions, thereby ensuring the accuracy of the collected data. At the same time, the system regularly updates semantic clusters through the preset dynamic theme library, which can adapt to changes in user focus and further improve the accuracy and timeliness of semantic recognition. On the other hand, the system determines the demand label based on preset quantitative parameters and user comment text, and stores the ternary mapping relationship between the product identification of the target product, the target theme and the demand label. This structured data storage method is not only convenient for subsequent analysis, but also can fully reflect the user's real needs and behavior patterns through multi-dimensional data fusion.

[0065] The present application also provides a computer-readable medium on which program instructions are stored. When the program instructions are executed by a processor, the data processing method based on the user's new media APP interactive operation provided by the above-mentioned method embodiments is implemented.

[0066] The present application also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the data processing method based on the user's new media APP interactive operation of the above-mentioned various method embodiments.

[0067] See also Figure 8 , is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 8 As shown, the electronic device 1000 may include: at least one processor 1001 , at least one network interface 1004 , a user interface 1003 , a memory 1005 , and at least one communication bus 1002 .

[0068] The communication bus 1002 is used to realize the connection and communication between these components.

[0069] The user interface 1003 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 1003 may also include a standard wired interface and a wireless interface.

[0070] The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0071] Among them, the processor 1001 may include one or more processing cores. The processor 1001 uses various interfaces and lines to connect various parts of the entire electronic device 1000, and executes various functions and processes data of the electronic device 1000 by running or executing instructions, programs, code sets or instruction sets stored in the memory 1005, and calling data stored in the memory 1005. Optionally, the processor 1001 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 1001 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 1001, and it can be implemented separately through a chip.

[0072] Among them, the memory 1005 may include a random access memory (Random Access Memory, RAM) and may also include a read-only memory (Read-Only Memory). Optionally, the memory 1005 includes a non-transitory computer-readable storage medium. The memory 1005 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 1005 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 1005 may also be optionally at least one storage system located away from the aforementioned processor 1001. As Figure 8 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a data processing application based on user new media APP interactive operations.

[0073] exist Figure 8In the electronic device 1000 shown, the user interface 1003 is mainly used to provide an input interface for the user and obtain the data input by the user; and the processor 1001 can be used to call the data processing application based on the user's new media APP interactive operation stored in the memory 1005, and specifically perform the following operations: In response to the user's interactive instruction on the comment box of the target object, the content input by the user on the comment box is obtained in real time; wherein the target object is the work browsed by the user in the new media APP, and the work is used to describe the target product; Performing contextual semantic analysis on the content of the real-time input comment box to identify whether there is a semantic cluster related to the content of the real-time input comment box in the semantic clusters corresponding to each theme in the preset dynamic theme library; wherein the preset dynamic theme library is regularly updated based on multiple original comments existing in the work within a preset period; In the case where there is a semantic cluster related to the content of the real-time input comment box, obtaining a target topic to which the semantic cluster related to the content of the real-time input comment box belongs, taking a regional screenshot of the comment box, and extracting all user comment texts in the regional screenshot; Determine the user's demand label for the target product based on the preset quantitative parameters and user comment text; In the database, a ternary mapping relationship between the product identification of the target product, the target subject and the demand label is stored.

[0074] In one embodiment, the processor 1001 further performs the following operations: Get multiple original comments on the work within a preset period; Analyze the original comments that are not related to the topics in the preset dynamic topic library from multiple original comments as the comments to be analyzed; Input each comment to be analyzed into a pre-trained semantic cluster analysis model, and output the semantic cluster corresponding to each comment to be analyzed; wherein the pre-trained semantic cluster analysis model is a data model obtained by machine learning based on historical user comment data stored in a preset time period; Classify the semantic clusters corresponding to each comment to be analyzed to obtain the topic corresponding to each comment to be analyzed; From the topics corresponding to each comment to be analyzed, count the number of topics with the same topic; When the number of topics of the same topic is greater than or equal to a preset number threshold, a binary mapping relationship between the same topic and all semantic clusters corresponding to the same topic is stored in a preset dynamic topic library.

[0075] In one embodiment, when the processor 1001 generates a pre-trained semantic cluster analysis model, the processor 1001 specifically performs the following operations: Collect historical user comment data stored in a preset time period from the new media APP, where the historical user comment data covers multiple works; Perform data preprocessing on historical user comment data to obtain the vocabulary sequence corresponding to each historical user comment data. Data preprocessing includes text cleaning, word segmentation, stop word removal, and part-of-speech tagging. Semantic information is captured for the word sequence corresponding to each piece of historical user comment data, so as to convert the word sequence corresponding to each piece of historical user comment data into a high-dimensional vector, and obtain multiple word features for each sample comment data; According to multiple lexical features of each sample comment data, a semantic cluster of each sample comment data is constructed; The semantic cluster of each sample comment data is used as a label, and the vocabulary sequence corresponding to each historical user comment data is annotated to obtain the model training sample; Create semantic cluster analysis model; Input the model training samples into the semantic cluster analysis model for machine learning and output the model loss value; When the model loss value reaches the minimum, a pre-trained semantic cluster analysis model is generated.

[0076] In one embodiment, when the processor 1001 constructs a semantic cluster of each sample comment data according to multiple vocabulary features of each sample comment data, the processor 1001 specifically performs the following operations: Initialize multiple lexical features of each sample comment data to mark the multiple lexical features as unvisited; Traversing the first unvisited lexical feature among the multiple lexical features of each sample comment data, and using the traversed lexical feature as the lexical feature to be analyzed, and marking the state of the lexical feature to be analyzed as visited; querying all the first unvisited lexical features in the domain of the lexical feature to be analyzed; Calculating the similarity between the lexical feature to be analyzed and each first lexical feature; The first lexical feature with a similarity greater than or equal to a preset similarity threshold and the lexical feature to be analyzed are grouped into one cluster to obtain a clustering cluster for each sample comment data; Continue to perform the step of traversing the first unvisited lexical feature among the multiple lexical features of each sample comment data until all the multiple lexical features are marked as visited, thereby obtaining multiple clusters of each sample comment data; Reversely transform the lexical features in each cluster of each sample comment data to obtain the lexical sequence of each cluster of each sample comment data; The vocabulary sequence of each cluster of each sample comment data is input into the preset large language model to analyze the representative vocabulary of each cluster of each sample comment data to obtain the semantic cluster of each sample comment data.

[0077] In one embodiment, when the processor 1001 inputs the model training sample into the semantic cluster analysis model for machine learning and outputs the model loss value, the processor 1001 specifically performs the following operations: The acquisition module obtains the semantic cluster of each sample comment data; The probability fitting module fits the first probability distribution of the semantic cluster of each sample comment data; The probability prediction module predicts the second probability distribution of each sample comment data; The cross entropy loss value calculation module calculates the cross entropy loss value of each sample comment data according to the first probability distribution and the second distribution probability; The model loss value calculation module calculates the cross entropy loss value of the model training sample according to the cross entropy loss value of each sample comment data as the model loss value; The loss value output module outputs the model loss value.

[0078] In one embodiment, when the processor 1001 identifies whether there is a semantic cluster related to the content of the real-time input comment box in the semantic clusters corresponding to each theme in the preset dynamic theme library, the processor 1001 specifically performs the following operations: From the preset dynamic theme library, traverse and obtain the semantic clusters corresponding to each theme; Extract the semantic feature vector of the content input into the comment box in real time as the center point vector; The embedding vectors of the semantic clusters corresponding to each topic are used as multiple edge vectors of the center point vector, and the relationship between the center point vector and each edge vector is established to obtain multiple objects to be analyzed; For each object to be analyzed, the semantic similarity between the center point vector and each edge vector is calculated to obtain multiple semantic similarities of each object to be analyzed; Calculate the similarity average of multiple semantic similarities of each object to be analyzed to obtain the target semantic similarity of each object to be analyzed; When there is an object to be analyzed whose target semantic similarity is greater than a preset threshold, it is determined that there is a semantic cluster related to the content of the real-time input comment box in the semantic clusters corresponding to each topic in the preset dynamic topic library; or When there is no object to be analyzed whose target semantic similarity is greater than a preset threshold, it is determined that there is no semantic cluster related to the content of the real-time input comment box in the semantic clusters corresponding to each topic in the preset dynamic topic library.

[0079] In one embodiment, when the processor 1001 determines the user's demand tag for the target product according to the preset quantitative parameters and the user comment text, the processor 1001 specifically performs the following operations: Use sentiment analysis tools to analyze the sentiment tendency of user comment texts and obtain the user's sentiment polarity score; Determine the user's emotional tendency description words based on the emotional threshold and the user's emotional polarity score; Match the user comment text with the preset keyword template through the TF-IDF algorithm to determine multiple keywords in the user comment text; Count the length of user review text as the user's attention to the target product; Determine the user's attention tendency description words according to the attention threshold and attention degree; Construct prompt words for the target product based on sentiment tendency descriptors, attention tendency descriptors, and multiple keywords. The prompt words are used to guide the preset large language model to generate user demand labels for the target product. Input the prompt words into the preset large language model and output the user's demand label for the target product; The default large language model is ChatGPT.

[0080] In one embodiment, the processor 1001 further performs the following operations: Receiving a data query request input by a client, the data query request carrying an identifier of a product to be analyzed and a subject to be analyzed; From the ternary mapping relationship, obtain all the required labels that satisfy the product identification to be analyzed and the subject to be analyzed; Aggregate all the same requirement tags to obtain non-repeated requirement tags; Import non-repeated requirement tags into the visualization component to visualize the non-repeated requirement tags.

[0081] In the embodiment of the present application, on the one hand, by acquiring user input content in real time and combining it with the contextual semantic analysis of each topic in the preset dynamic theme library, the system can identify semantic clusters in user comments in real time, not just keywords. This semantic-based analysis method can effectively reduce misjudgments caused by the diversity of language expressions, thereby ensuring the accuracy of the collected data. At the same time, the system regularly updates semantic clusters through the preset dynamic theme library, which can adapt to changes in user focus and further improve the accuracy and timeliness of semantic recognition. On the other hand, the system determines the demand label based on preset quantitative parameters and user comment text, and stores the ternary mapping relationship between the product identification of the target product, the target theme and the demand label. This structured data storage method is not only convenient for subsequent analysis, but also can fully reflect the user's real needs and behavior patterns through multi-dimensional data fusion.

[0082] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program for data processing based on the interactive operation of the user's new media APP can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium of the program for data processing based on the interactive operation of the user's new media APP can be a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0083] The above disclosure is only the preferred embodiment of the present application, which certainly cannot be used to limit the scope of rights of the present application. Therefore, equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A data processing method based on user new media APP interactive operation, characterized in that: The method comprises: In response to an interactive instruction of a user on a comment box of a target object, obtaining content input by the user on the comment box in real time; wherein the target object is a work browsed by the user in a new media APP, and the work is used to describe a target product; Performing contextual semantic analysis on the content input into the comment box in real time to identify whether there is a semantic cluster related to the content input into the comment box in real time in the semantic clusters corresponding to each theme in the preset dynamic theme library; wherein the preset dynamic theme library is regularly updated based on a plurality of original comments existing in the work within a preset period; In the case where there is a semantic cluster related to the content input into the comment box in real time, obtaining a target topic to which the semantic cluster related to the content input into the comment box in real time belongs, taking a regional screenshot of the comment box, and extracting all user comment texts in the regional screenshot; Determining the user's demand label for the target product according to the preset quantitative parameters and the user comment text; In the database, a ternary mapping relationship between the product identification of the target product, the target subject and the demand tag is stored.

2. The method according to claim 1, characterized in that The method further comprises: Obtain multiple original reviews of the work within a preset period; Analyzing the original comments that are not related to the topics in the preset dynamic topic library from the multiple original comments as the comments to be analyzed; Input each comment to be analyzed into a pre-trained semantic cluster analysis model, and output the semantic cluster corresponding to each comment to be analyzed; wherein the pre-trained semantic cluster analysis model is a data model obtained by machine learning based on historical user comment data stored in a preset time period; Performing topic classification on the semantic cluster corresponding to each comment to be analyzed to obtain the topic corresponding to each comment to be analyzed; From the topics corresponding to each of the comments to be analyzed, count the number of topics with the same topic; When the number of topics of the same topic is greater than or equal to a preset number threshold, a binary mapping relationship between the same topic and all semantic clusters corresponding to the same topic is stored in the preset dynamic topic library.

3. The method according to claim 2, characterized in that Follow these steps to generate a pre-trained semantic clustering analysis model, including: Collecting historical user comment data stored in a preset time period from the new media APP, where the historical user comment data covers multiple works; Performing data preprocessing on the historical user comment data to obtain a vocabulary sequence corresponding to each piece of historical user comment data, wherein the data preprocessing includes text cleaning, word segmentation, stop word removal, and part-of-speech tagging; Capturing semantic information of the vocabulary sequence corresponding to each piece of historical user comment data, so as to convert the vocabulary sequence corresponding to each piece of historical user comment data into a high-dimensional vector, and obtaining a plurality of vocabulary features of each piece of sample comment data; Constructing a semantic cluster of each piece of sample comment data according to a plurality of lexical features of each piece of sample comment data; Using the semantic cluster of each sample comment data as a label, annotating the vocabulary sequence corresponding to each historical user comment data, and obtaining a model training sample; Create semantic cluster analysis model; Inputting the model training samples into the semantic cluster analysis model for machine learning, and outputting a model loss value; When the model loss value reaches the minimum, a pre-trained semantic cluster analysis model is generated.

4. The method according to claim 3, characterized in that The step of constructing a semantic cluster of each piece of sample comment data according to the plurality of lexical features of each piece of sample comment data comprises: Initializing a plurality of lexical features of each sample comment data to mark the plurality of lexical features as unvisited; Traversing the first unvisited lexical feature among the multiple lexical features of each sample comment data, taking the traversed lexical feature as the lexical feature to be analyzed, and marking the state of the lexical feature to be analyzed as visited; Querying all unvisited first vocabulary features in the field of the vocabulary feature to be analyzed; Calculating the similarity between the to-be-analyzed vocabulary feature and each first vocabulary feature; Classifying the first vocabulary feature whose similarity is greater than or equal to a preset similarity threshold and the vocabulary feature to be analyzed into one cluster, to obtain a clustering cluster for each sample comment data; Continue to perform the step of traversing the first unvisited lexical feature among the multiple lexical features of each sample comment data until all the multiple lexical features are marked as visited, thereby obtaining multiple clusters of each sample comment data; Reversely transforming the vocabulary features in each cluster of each sample comment data to obtain the vocabulary sequence of each cluster of each sample comment data; The vocabulary sequence of each cluster of each sample comment data is input into a preset large language model to analyze the representative vocabulary of each cluster of each sample comment data to obtain the semantic cluster of each sample comment data.

5. The method according to claim 3, characterized in that: The model training sample includes the semantic cluster of each sample comment data and each sample comment data; the semantic cluster analysis model includes an acquisition module, a probability fitting module, a probability prediction module, a cross entropy loss value calculation module, a model loss value calculation module and a loss value output module; The step of inputting the model training sample into the semantic cluster analysis model for machine learning and outputting the model loss value comprises: The acquisition module acquires the semantic cluster of each sample comment data; The probability fitting module fits a first probability distribution of a semantic cluster of each sample comment data; The probability prediction module predicts a second probability distribution of each sample comment data; The cross entropy loss value calculation module calculates the cross entropy loss value of each sample comment data according to the first probability distribution and the second distribution probability; The model loss value calculation module calculates the cross entropy loss value of the model training sample according to the cross entropy loss value of each sample comment data as the model loss value; The loss value output module outputs the model loss value.

6. The method according to claim 1, characterized in that The step of identifying whether there is a semantic cluster related to the content input into the comment box in real time in the semantic clusters corresponding to each theme in the preset dynamic theme library, includes: From the preset dynamic theme library, traverse and obtain the semantic clusters corresponding to each theme; Extracting a semantic feature vector of the content input into the comment box in real time as a center point vector; Using the embedding vectors of the semantic clusters corresponding to the topics as multiple edge vectors of the center point vector, and establishing a relationship between the center point vector and each edge vector to obtain multiple objects to be analyzed; For each object to be analyzed, calculating the semantic similarity between the center point vector and each edge vector to obtain multiple semantic similarities of each object to be analyzed; Calculating a similarity mean of multiple semantic similarities of each object to be analyzed to obtain a target semantic similarity of each object to be analyzed; When there is an object to be analyzed whose target semantic similarity is greater than a preset threshold, it is determined that there is a semantic cluster related to the content input into the comment box in real time among the semantic clusters corresponding to each topic in the preset dynamic topic library; or, when there is no object to be analyzed whose target semantic similarity is greater than the preset threshold, it is determined that there is no semantic cluster related to the content input into the comment box in real time among the semantic clusters corresponding to each topic in the preset dynamic topic library.

7. The method according to claim 1, characterized in that The preset quantitative parameters include an emotion threshold and an attention threshold; The step of determining the user's demand label for the target product according to the preset quantitative parameter and the user comment text includes: Performing sentiment analysis on the user's comment text to obtain the user's sentiment polarity score; Determining a descriptive word for the user's emotional tendency according to the emotional threshold and the user's emotional polarity score; Matching the user comment text with a preset keyword template by using a TF-IDF algorithm to determine a plurality of keywords in the user comment text; Counting the length of the user's comment text as the user's attention to the target product; Determining a descriptive word for the user's attention tendency according to the attention threshold and the attention degree; Constructing a prompt word for the target product according to the sentiment tendency description word, the attention tendency description word and the multiple keywords, wherein the prompt word is used to guide a preset large language model to generate a demand label of the user for the target product; The prompt word is input into a preset large language model, and a demand label of the user for the target product is output.

8. The method according to claim 1, characterized in that After storing the ternary mapping relationship between the product identification of the target product, the target subject and the demand tag, the method further includes: Receiving a data query request input by a client, wherein the data query request carries an identification of a product to be analyzed and a subject to be analyzed; From the ternary mapping relationship, obtain all the required tags that satisfy the product identifier to be analyzed and the subject to be analyzed; Aggregate all the same requirement tags to obtain non-repeated requirement tags; Import non-repeated requirement tags into the visualization component to visualize the non-repeated requirement tags.

9. A data processing device based on user new media APP interactive operation, characterized in that: The device comprises: A response acquisition module, for responding to an interactive instruction of a user on a comment box of a target object, and acquiring in real time the content input by the user on the comment box; wherein the target object is a work browsed by the user in a new media APP, and the work is used to describe a target product; A semantic cluster determination module is used to perform contextual semantic analysis on the content input into the comment box in real time, so as to identify whether there is a semantic cluster related to the content input into the comment box in real time in the semantic clusters corresponding to each theme in the preset dynamic theme library; wherein the preset dynamic theme library is obtained by regularly updating a plurality of original comments existing in the work within a preset period; A text extraction module, for obtaining, when there is a semantic cluster related to the content input into the comment box in real time, a target topic to which the semantic cluster related to the content input into the comment box in real time belongs, taking a regional screenshot of the comment box, and extracting all user comment texts in the regional screenshot; A demand label determination module, used to determine the user's demand label for the target product according to preset quantitative parameters and the user comment text; The ternary mapping relationship storage module is used to store the ternary mapping relationship between the product identification of the target product, the target subject and the demand tag in a database.

Images