Method for discovering network information topics and their concerned user groups

Abstract

The present invention discloses a method for discovering network information topics and user groups concerned therewith, and relates to the technical field of user groups concerned with network information, in order to solve the problem of unclear information data topic analysis and inaccurate user group discovery. The present invention can more accurately understand the inherent relationship and characteristics of users and information data through a series of steps such as curve data conversion, sub-curve data construction, overlapping comparison, and judgment of the proportion of concerned groups, and provide more valuable data insights and analysis results. By extracting concerned browsing users in target information data and further analyzing the characteristics of users, it can provide more in-depth user insights and more targeted data support. By extracting the topic features of information data, it can better understand and represent the topic information and semantic information of the data, and provide more accurate and representative feature input for subsequent topic modeling.

Description

Technical Field

[0001] The present invention relates to the technical field of network information attention user groups, and in particular to a method for discovering network information topics and attention user groups thereof. Background Art

[0002] The group of users who pay attention to online information refers to the group of people who pay attention to a certain information event on the Internet and express their views, opinions, comments, etc.

[0003] The Chinese patent with publication number CN115905724A discloses a method for tracing the source of major information events based on a double-layer information blockchain. It mainly achieves the full-process traceability of each user behavior in the public opinion platform and the social network platform by constructing a double-layer blockchain, based on the blockchain's tamper-proof and traceability, and realizes the rapid identification of abnormal information by integrating the growth rate of user node influence and the sentiment analysis algorithm. Through the double-layer blockchain, the user behavior and the user's community data are queried, and the information of individuals and groups that promote the development of information is mined to achieve the traceability of major information events. Although the above patent solves the problem of information tracing, there are still the following problems in actual operation:

[0004] 1. Failure to conduct a more accurate analysis of the relationship between information data and the users being followed, resulting in a lack of understanding of the relationship between information data and the user groups being followed.

[0005] 2. Failure to conduct a more detailed analysis of the characteristics of users browsing information data results in an inability to more accurately understand the behavioral connections between different users and different information data.

[0006] 3. The network information data is not divided into detailed topics according to the content of the data, resulting in unclear themes of the information data. Summary of the invention

[0007] The purpose of the present invention is to provide a method for discovering network information topics and their concerned user groups. Through a series of steps such as curve data conversion, sub-curve data construction, overlapping comparison and judgment of the proportion of concerned groups, the intrinsic relationship and characteristics of users and information data can be more accurately understood, and more valuable data insights and analysis results can be provided. By extracting the concerned browsing users in the target information data and further analyzing the characteristics of the users, deeper user insights and more targeted data support can be provided. The subject features of the information data are extracted to better understand and represent the subject information and semantic information of the data, providing more accurate and representative feature input for subsequent subject modeling. The subject model of the information data is constructed according to the subject features, thereby improving the accuracy of the construction of the subject model of the information data, improving the intelligence level and application value of data processing, and solving the problems in the prior art.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] The method for discovering network information topics and their concerned user groups includes the following steps:

[0010] S1: Network information data collection: network information data is acquired through different acquisition platforms, and the acquired data and the platform where the data is acquired are coded accordingly;

[0011] S2: Network information data analysis: pre-process the acquired data according to the corresponding coding, perform topic modeling on the acquired data after the data pre-processing, and obtain the target information data after topic modeling;

[0012] S3: Information collection of concerned user information: confirm the concerned user information in the target information data, extract user portrait features based on the concerned user information, and mark the extracted user screen features as standard user data;

[0013] S4: User information feature analysis: Overlap the features of standard user data and target information data, and determine the proportion of the target information data’s attention group based on the overlapping area.

[0014] Preferably, the network information data in S1 is acquired through different acquisition platforms, including:

[0015] The platforms for obtaining online information data include social media platforms, website optimization platforms, third-party data platforms, and manual retrieval platforms;

[0016] Match the information data obtained from different assessments with the platform respectively;

[0017] After the correspondence is completed, the acquired information data and the unique coding number of the acquisition platform are obtained.

[0018] Preferably, data preprocessing is performed on the data to be acquired in S2, including:

[0019] Confirm the information data in the unique coding number;

[0020] After confirmation, the information data will be cleaned, normalized and converted in sequence;

[0021] Among them, data cleaning first processes missing values, abnormal values ​​and duplicate values ​​of information data, and obtains first processed data after processing;

[0022] Then scaling the first processed data to within a preset range, completing data normalization processing after scaling, and obtaining second processed data;

[0023] The second processed data is converted into a modeling format, and after the conversion, the data conversion process is completed to obtain the third processed data.

[0024] Preferably, the data cleaning process includes:

[0025] Extracting text data information from the information data;

[0026] Performing text recognition on the text data information to determine whether there is a garbled field in the text data information;

[0027] When there is a garbled field in the text data information, a weight parameter corresponding to the garbled field is obtained according to the parameter information of the garbled field; wherein the weight parameter is obtained by the following formula:

[0028]

[0029] Where W represents the weight parameter corresponding to the garbled field; F m Indicates the number of characters corresponding to the garbled field; F j Indicates the number of characters corresponding to the short sentence in which the garbled field is located; F d represents the number of characters corresponding to the paragraph in the paragraph where the garbled field is located; λ represents the cosine similarity between the average word vector corresponding to the upper part of the sentence information of the garbled field and the average word vector corresponding to the lower part of the sentence information;

[0030] When the weight parameter of the garbled code field is lower than a preset weight threshold, it is determined whether the garbled code field needs to be deleted;

[0031] When the weight parameter of the garbled code field is not lower than a preset weight threshold, the original data information of the text data information corresponding to the garbled code part is retrieved from the data interface end.

[0032] Preferably, when the weight parameter of the garbled code field is lower than a preset weight threshold, determining whether the garbled code field needs to be deleted includes:

[0033] When the weight parameter of the garbled code field is lower than a preset weight threshold, the upper sentence information and the lower sentence information of the text part where the garbled code field is located are retrieved;

[0034] Obtaining the Euclidean distance between the average word vector corresponding to the upper part of the sentence information and the average word vector corresponding to the lower part of the sentence information;

[0035] The weight parameter of the garbled field is compensated by using the Euclidean distance between the average word vector corresponding to the upper part of the sentence information and the average word vector corresponding to the lower part of the sentence information to obtain the weight parameter after compensation; wherein the weight parameter after compensation is obtained by the following formula:

[0036]

[0037] Among them, W b represents the weight parameter after compensation processing; O represents the Euclidean distance between the average word vector corresponding to the upper part of the sentence information and the average word vector corresponding to the lower part of the sentence information; O 0 represents the preset Euclidean distance threshold;

[0038] When the weight parameter after the compensation process is lower than the preset weight threshold, the garbled field is deleted;

[0039] When the weight parameter after the compensation process is not lower than the preset weight threshold, the original data information of the text data information corresponding to the garbled part is retrieved from the data interface end.

[0040] Preferably, topic modeling is performed on the data acquired in S2, including:

[0041] Before subject modeling is performed on the third processed data, data subject features are first extracted from the third processed data;

[0042] Among them, first confirm the data subject type of the third processed data, which includes political information subject, economic information subject, cultural information subject and social information subject;

[0043] Performing data theme analysis on the third processed data, dividing the third processed data into a multi-layer theme tree, each layer including a plurality of clusters;

[0044] Cluster the sub-data information corresponding to each layer to obtain several cluster sets, and assign each cluster set to the corresponding cluster;

[0045] Performing a word segmentation operation on the sentence type of the third processed data included in the cluster set in each cluster to obtain a plurality of extracted words in the third processed data;

[0046] Determine the word feature of each extracted word and the sentence feature of each extracted word in the sentence type.

[0047] Preferably, performing topic modeling on the data acquired in S2 further includes:

[0048] Determine the similarity between the extracted words based on the constructed synonym dictionary and the word features of each extracted word;

[0049] Obtain a word set of extracted words according to the similarity percentage;

[0050] The extracted words with consecutive positions in the word set are concatenated with the maximum length in the smallest text unit to obtain concatenated words;

[0051] Clean the concatenated words according to the statistical features of parts of speech to obtain a keyword set;

[0052] Convert keywords in the keyword set into word vectors;

[0053] Confirm the distance between each word vector in any cluster and the standard word vector corresponding to each cluster, and select the keyword corresponding to the word vector with the smallest distance as the target keyword in any cluster;

[0054] The keywords included in each layer of the topic tree are determined according to the target keywords in any cluster.

[0055] Preferably, performing topic modeling on the data acquired in S2 further includes:

[0056] Match the corresponding information topic samples in the data topic type according to the keywords;

[0057] Determine the data classification identifier based on the information subject sample, and at the same time, input the data classification identifier into a preset neural network for learning to determine the classification expression of the data classification identifier;

[0058] An information topic classification model is constructed according to the classification expression, and at the same time, the information topic samples are input into the information topic classification model for classification;

[0059] After the classification is completed, the subject type of the third processed data and the modeling model of the subject type are obtained;

[0060] The topic type of the third processed data and the modeling model of the topic type are labeled as target information data.

[0061] Preferably, extracting user portrait features based on the concerned user information in S3 includes:

[0062] Retrieve the concerned browsing users in the target information data;

[0063] The browsing time data of the followed browsing users are confirmed, and the followed browsing users whose browsing time is within a preset range are taken as target followed browsing users;

[0064] Retrieve the browsing IP address of the target browsing user;

[0065] Confirm the target user profile based on the browsing IP address;

[0066] Among them, user portrait features include the region, age, and gender of the target browsing user;

[0067] And mark the confirmed user portrait features as standard user data.

[0068] Preferably, judging the proportion of the concerned group of the target information data according to the overlapping area in S4 includes:

[0069] Convert standard user data and target information data into curve data respectively;

[0070] After the curve data conversion is completed, the curve user data and curve information data are obtained;

[0071] Then, the curve user data is used to construct sub-curve data according to the user portrait characteristics;

[0072] Compare the curve information data with the sub-curve data for curve overlap;

[0073] After the curves overlap, the proportion of the concerned groups of the curve information data is determined according to the overlap rate of the overlap area;

[0074] The higher the overlap degree between the curve information data and the sub-curve data, the higher the overlap rate between the curve information data and the sub-curve data, and the closer the focus group of the curve information data is to the sub-curve data.

[0075] Compared with the prior art, the present invention has the following beneficial effects:

[0076] 1. The method for discovering network information topics and their concerned user groups provided by the present invention extracts the topic features of the information data after the information data preprocessing is completed. Extracting the topic features of the information data can better understand and represent the topic information and semantic information of the data, and provide more accurate and representative feature input for subsequent topic modeling. The topic model of the information data is constructed according to the topic features, thereby improving the accuracy of the information data topic model construction, improving the intelligent level of data processing and the application value, which helps to better understand and analyze the information data and provide strong data support for related decisions and applications.

[0077] 2. The method for discovering network information topics and their concerned user groups provided by the present invention can provide deeper user insights and more targeted data support by extracting concerned browsing users from target information data and further analyzing user characteristics, which helps to better understand user needs and behaviors and enhance the application value and service level of data.

[0078] 3. The method for discovering network information topics and their concerned user groups provided by the present invention can more accurately understand the intrinsic relationship and characteristics of users and information data through a series of steps such as curve data conversion, sub-curve data construction, overlapping comparison and judgment of the proportion of concerned groups, and provide more valuable data insights and analysis results, which helps to better understand users' interests and behaviors, as well as the dissemination and influence of information data, and provide more targeted data support for related decisions and applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 It is a schematic diagram of the overall method flow of the present invention;

[0080] Figure 2 It is a schematic diagram of obtaining network information data of the present invention;

[0081] Figure 3 A schematic diagram of user data acquisition of the present invention;

[0082] Figure 4 It is a schematic diagram of the proportion of interest of the present invention. DETAILED DESCRIPTION

[0083] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0084] In order to solve the problem in the prior art that after collecting network information data, the network information data is not divided into detailed topics according to the content of the data, resulting in unclear topics of the information data, please refer to Figure 1-Figure 2 , this embodiment provides the following technical solutions:

[0085] The method for discovering network information topics and their concerned user groups includes the following steps:

[0086] S1: Network information data collection: network information data is acquired through different acquisition platforms, and the acquired data and the platform where the data is acquired are coded accordingly;

[0087] Among them, acquiring different network information data according to different acquisition platforms can further improve the contrast of data, so as to have a more comprehensive understanding of different information situations;

[0088] S2: Network information data analysis: pre-process the acquired data according to the corresponding coding, perform topic modeling on the acquired data after the data pre-processing, and obtain the target information data after topic modeling;

[0089] Among them, the topic model of information data is constructed according to the topic characteristics, thereby improving the accuracy of the information data topic model construction, and improving the intelligent level and application value of data processing;

[0090] S3: Information collection of concerned user information: confirm the concerned user information in the target information data, extract user portrait features based on the concerned user information, and mark the extracted user screen features as standard user data;

[0091] The construction of user portraits can provide deeper user insights, help understand user behaviors and preferences, and identify target browsing users through a preset browsing time range, which can ensure the integrity and accuracy of the data.

[0092] S4: User information feature analysis: Overlap the standard user data and target information data, and determine the proportion of the target information data’s attention group based on the overlapping area;

[0093] Among them, judging the proportion of attention groups of the curve information data according to the overlapping rate of the overlapping area can more accurately understand the degree of attention and influence of different user groups on specific information topics.

[0094] The network information data in S1 is obtained through different acquisition platforms, including:

[0095] The platforms for obtaining online information data include social media platforms, website optimization platforms, third-party data platforms, and manual retrieval platforms;

[0096] Match the information data obtained from different assessments with the platform respectively;

[0097] After the correspondence is completed, the acquired information data and the unique coding number of the acquisition platform are obtained.

[0098] Specifically, acquiring different network information data according to different acquisition platforms can further improve the contrast of the data, so that different information situations can be more comprehensively understood. At the same time, acquiring information data through different platforms can verify the information data between platforms, thereby improving the reliability and credibility of the network information data.

[0099] Perform data preprocessing on the data to be acquired in S2, including:

[0100] Confirm the information data in the unique coding number;

[0101] After confirmation, the information data will be cleaned, normalized and converted in sequence;

[0102] Among them, data cleaning first processes missing values, abnormal values ​​and duplicate values ​​of information data, and obtains first processed data after processing;

[0103] Then scaling the first processed data to within a preset range, completing data normalization processing after scaling, and obtaining second processed data;

[0104] The second processed data is converted into a modeling format, and after the conversion, the data conversion process is completed to obtain the third processed data.

[0105] Specifically, preprocessing the data after acquiring the information can improve the quality, standardization and processing efficiency of the data, and provide a more reliable and accurate data basis for subsequent data analysis and model training. Among them, processing the data for missing values, outliers and duplicate values ​​can make the data analysis results more accurate and reliable. The first processed data is normalized, and the data is first scaled to a preset range to complete the normalization of the data, which is convenient for subsequent data analysis and processing. By converting the modeling format, the data can be converted into a form suitable for model training and prediction, thereby improving the efficiency and accuracy of data analysis.

[0106] At the same time, the data cleaning process includes:

[0107] Extracting text data information from the information data;

[0108] Performing text recognition on the text data information to determine whether there is a garbled field in the text data information;

[0109] When there is a garbled field in the text data information, a weight parameter corresponding to the garbled field is obtained according to the parameter information of the garbled field; wherein the weight parameter is obtained by the following formula:

[0110]

[0111] Where W represents the weight parameter corresponding to the garbled field; F m Indicates the number of characters corresponding to the garbled field; F j Indicates the number of characters corresponding to the short sentence in which the garbled field is located; F d represents the number of characters corresponding to the paragraph in the paragraph where the garbled field is located; λ represents the cosine similarity between the average word vector corresponding to the upper part of the sentence information of the garbled field and the average word vector corresponding to the lower part of the sentence information;

[0112] When the weight parameter of the garbled code field is lower than a preset weight threshold, it is determined whether the garbled code field needs to be deleted;

[0113] When the weight parameter of the garbled code field is not lower than a preset weight threshold, the original data information of the text data information corresponding to the garbled code part is retrieved from the data interface end.

[0114] The technical effect of the above technical solution is: by cleaning the text data in the information data, the garbled fields therein can be eliminated, thereby significantly improving the overall quality of the data. The presence of garbled data may lead to misjudgment or failure of subsequent analysis, so the cleaning process is crucial to ensure the validity of the data. Through text recognition technology, the technical solution can accurately determine whether there are garbled fields in the text data information. This avoids blindly deleting or modifying data and ensures the accuracy of data processing.

[0115] The processing of garbled fields is not a one-size-fits-all approach, but is judged based on its weight parameters. This processing method takes into account the importance of the garbled fields in the context, helps to retain key information, and avoids accidentally deleting important data. The calculation of the weight parameter comprehensively considers the number of characters in the garbled field, the number of characters in the short sentence and paragraph, and the semantic similarity of the contextual sentence information. This comprehensive evaluation method can more comprehensively reflect the actual impact of the garbled field and improve the rationality of the decision. For garbled fields whose weight parameters are lower than the preset threshold, the technical solution provides a deletion option to reduce data noise. For fields with higher weight parameters, the original data information is retrieved to try to restore or correct them, thereby retaining more valuable information. The above technical solution of this embodiment uses word vector technology to calculate the semantic similarity of the context of the garbled field, which helps to more accurately evaluate the impact of the garbled field in the overall data and enhance the accuracy and pertinence of data cleaning.

[0116] In summary, the above technical solution of this embodiment effectively improves the quality of information data through a scientific and accurate data cleaning process, and provides a more reliable basis for subsequent data analysis and decision-making.

[0117] Specifically, when the weight parameter of the garbled code field is lower than a preset weight threshold, determining whether the garbled code field needs to be deleted includes:

[0118] When the weight parameter of the garbled code field is lower than a preset weight threshold, the upper sentence information and the lower sentence information of the text part where the garbled code field is located are retrieved;

[0119] Obtaining the Euclidean distance between the average word vector corresponding to the upper part of the sentence information and the average word vector corresponding to the lower part of the sentence information;

[0120] The weight parameter of the garbled field is compensated by using the Euclidean distance between the average word vector corresponding to the upper part of the sentence information and the average word vector corresponding to the lower part of the sentence information to obtain the weight parameter after compensation; wherein the weight parameter after compensation is obtained by the following formula:

[0121]

[0122] Among them, W b represents the weight parameter after compensation processing; O represents the Euclidean distance between the average word vector corresponding to the upper part of the sentence information and the average word vector corresponding to the lower part of the sentence information; O 0 represents the preset Euclidean distance threshold;

[0123] When the weight parameter after the compensation process is lower than the preset weight threshold, the garbled field is deleted;

[0124] When the weight parameter after the compensation process is not lower than the preset weight threshold, the original data information of the text data information corresponding to the garbled part is retrieved from the data interface end.

[0125] The technical effect of the above technical solution is that when judging whether the garbled field needs to be deleted, the Euclidean distance between the average word vectors of the upper and lower sentence information is introduced as a compensation parameter. This processing method takes into account the semantic information of the context of the garbled field, making the judgment more accurate.

[0126] At the same time, traditional data cleaning methods may directly delete garbled fields based on weight parameters, but this method may lead to the loss of important information. By introducing compensation processing, even if the original weight parameter of the garbled field is low, if its contextual semantics is similar to the overall text, it may be retained, reducing the risk of accidental deletion.

[0127] On the other hand, for the garbled fields whose weight parameters after compensation are still not lower than the preset threshold, the above technical solution chooses to retrieve the original data information from the data interface end and try to restore or correct the garbled part. This can retain the original information of the data to the greatest extent and improve the integrity of the data. In addition, the weight threshold and Euclidean distance threshold in the above technical solution can be flexibly adjusted according to actual needs to meet the data cleaning needs in different scenarios. This technical solution not only takes into account the characteristics of the garbled field itself (such as the number of characters, the number of characters in the short sentence and paragraph, etc.), but also combines the semantic similarity of the context sentence information, as well as the Euclidean distance between the garbled field and the context, and makes a comprehensive consideration to make data cleaning more scientific and reasonable.

[0128] In summary, the above technical solution improves the accuracy and completeness of data cleaning, reduces the risk of accidental deletion, and makes the data cleaning process more flexible and scientific by introducing Euclidean distance as a compensation parameter. This helps to improve the quality of information data and provide a more reliable basis for subsequent data analysis and decision-making.

[0129] Perform topic modeling on the data to be obtained in S2, including:

[0130] Before subject modeling is performed on the third processed data, data subject features are first extracted from the third processed data;

[0131] Among them, first confirm the data subject type of the third processed data, which includes political information subject, economic information subject, cultural information subject and social information subject;

[0132] Performing data theme analysis on the third processed data, dividing the third processed data into a multi-layer theme tree, each layer including a plurality of clusters;

[0133] Cluster the sub-data information corresponding to each layer to obtain several cluster sets, and assign each cluster set to the corresponding cluster;

[0134] Performing a word segmentation operation on the sentence type of the third processed data included in the cluster set in each cluster to obtain a plurality of extracted words in the third processed data;

[0135] Determine the word feature of each extracted word and the sentence feature of each extracted word in the sentence type.

[0136] Determine the similarity between the extracted words based on the constructed synonym dictionary and the word features of each extracted word;

[0137] Obtain a word set of extracted words according to the similarity percentage;

[0138] The extracted words with consecutive positions in the word set are concatenated with the maximum length in the smallest text unit to obtain concatenated words;

[0139] Clean the concatenated words according to the statistical features of parts of speech to obtain a keyword set;

[0140] Convert keywords in the keyword set into word vectors;

[0141] Confirm the distance between each word vector in any cluster and the standard word vector corresponding to each cluster, and select the keyword corresponding to the word vector with the smallest distance as the target keyword in any cluster;

[0142] The keywords included in each layer of the topic tree are determined according to the target keywords in any cluster.

[0143] Match the corresponding information topic samples in the data topic type according to the keywords;

[0144] Determine the data classification identifier based on the information subject sample, and at the same time, input the data classification identifier into a preset neural network for learning to determine the classification expression of the data classification identifier;

[0145] An information topic classification model is constructed according to the classification expression, and at the same time, the information topic samples are input into the information topic classification model for classification;

[0146] After the classification is completed, the subject type of the third processed data and the modeling model of the subject type are obtained;

[0147] The topic type of the third processed data and the modeling model of the topic type are labeled as target information data.

[0148] Specifically, after the preprocessing of the information data is completed, the topic features of the information data are extracted. Extracting the topic features of the information data can better understand and represent the topic information and semantic information of the data, and provide more accurate and representative feature inputs for subsequent topic modeling. This helps to improve the accuracy and reliability of topic modeling, and enhance the intelligence and interpretability of data analysis. Among them, by confirming the data topic type of the third processed data, the data can be focused on specific fields such as politics, economy, culture and society, making the subsequent topic modeling more accurate and targeted, and dividing the third processed data into a multi-layer topic tree, each layer includes multiple clusters. This hierarchical data structure can better reveal the topic relationship and hierarchy in the data. The sentence type of the third processed data included in the cluster set in each cluster is segmented, and several extracted words in the third processed data can be obtained. This word segmentation operation can better understand the semantic information of the data, which is helpful for subsequent feature representation and modeling. According to the constructed synonym dictionary and the word features of each extracted word, the similarity between the extracted words is determined. This synonym matching can enhance the diversity of features and the richness of semantics, and improve the generalization ability and accuracy of the model. After the topic feature extraction of information data is completed, the topic model of the information data is constructed according to the topic features, thereby improving the accuracy of the information data topic model construction, improving the intelligence level and application value of data processing, which helps to better understand and analyze information data and provide strong data support for related decisions and applications. Among them, the keywords in the keyword set are converted into word vectors. This word vector representation can better capture the semantic information and correlation between keywords, and provide a richer and more accurate feature representation for subsequent topic modeling. By calculating the distance between the word vector and the standard word vector, the keyword corresponding to the word vector with the smallest distance is selected as the target keyword in any cluster. This screening method can ensure that the selected keywords are most similar to the standard keywords, thereby improving the accuracy and reliability of topic modeling. By inputting the data classification identifier into the preset topic model for learning, the classification expression of the data classification identifier can be determined. According to the topic type of the third processed data and the modeling model of the topic type, it can be used to classify and predict new information data, thereby improving the automation and intelligence level of data analysis and processing.

[0149] In order to solve the problem in the prior art that the browsing user characteristics of information data are not analyzed in more detail, resulting in an inability to more accurately understand the behavioral connection between different users and different information data, please refer to Figure 3 , this embodiment provides the following technical solutions:

[0150] User portrait feature extraction based on user information in S3, including:

[0151] Retrieve the concerned browsing users in the target information data;

[0152] The browsing time data of the followed browsing users are confirmed, and the followed browsing users whose browsing time is within a preset range are taken as target followed browsing users;

[0153] Retrieve the browsing IP address of the target browsing user;

[0154] Confirm the target user profile based on the browsing IP address;

[0155] Among them, user portrait features include the region, age, and gender of the target browsing user;

[0156] And mark the confirmed user portrait features as standard user data.

[0157] Specifically, by extracting the focused browsing users from the target information data and further analyzing the user's characteristics, we can provide deeper user insights and more targeted data support, which helps to better understand the user's needs and behaviors and improve the application value and service level of the data. Among them, by retrieving the browsing IP of the target focused browsing users, we can confirm the user's portrait characteristics, including region, age, gender, etc. This kind of user portrait construction can provide deeper user insights and help understand user behavior and preferences. Confirming the target focused browsing users through a preset browsing time range can ensure the integrity and accuracy of the data and avoid affecting the accuracy of the user portrait due to too long or too short browsing time. Through the construction and analysis of user portraits, we can better understand the user's needs and behaviors and provide strong data support for the optimization of related products and services.

[0158] In order to solve the problem that the relationship between information data and the concerned users is not analyzed more accurately in the prior art, which leads to unclear understanding of the relationship between information data and the concerned user groups, please refer to Figure 4 , this embodiment provides the following technical solutions:

[0159] The proportion of the concerned groups for the target information data determined based on the overlapping areas in S4 includes:

[0160] Convert standard user data and target information data into curve data respectively;

[0161] After the curve data conversion is completed, the curve user data and curve information data are obtained;

[0162] Then, the curve user data is used to construct sub-curve data according to the user portrait characteristics;

[0163] Compare the curve information data with the sub-curve data for curve overlap;

[0164] After the curves overlap, the proportion of the concerned groups of the curve information data is determined according to the overlap rate of the overlap area;

[0165] The higher the overlap degree between the curve information data and the sub-curve data, the higher the overlap rate between the curve information data and the sub-curve data, and the closer the focus group of the curve information data is to the sub-curve data.

[0166] Specifically, through a series of steps such as curve data conversion, sub-curve data construction, overlap comparison and judgment of the proportion of attention groups, we can more accurately understand the intrinsic relationship and characteristics of users and information data, and provide more valuable data insights and analysis results, which is helpful to better understand users' interests and behaviors, as well as the dissemination and influence of information data, and provide more targeted data support for related decisions and applications. Among them, through curve data conversion, the originally discrete data can be converted into a continuous curve form, making the data representation more intuitive and visual. Sub-curve data construction of curve user data based on user portrait characteristics can further segment user groups and provide more accurate user insights. Curve overlap comparison of curve information data and sub-curve data can better identify the correlation and consistency between curve information data and different user groups, and can more accurately judge the proportion of attention groups of curve information data. Judging the proportion of attention groups of curve information data based on the overlap rate of overlapping areas can more accurately understand the degree of attention and influence of different user groups on specific information topics.

[0167] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0168] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for discovering network information topics and their concerned user groups, characterized in that: The steps include: S1: Network information data collection: network information data is acquired through different acquisition platforms, and the acquired data and the platform where the data is acquired are coded accordingly; S2: Network information data analysis: pre-process the acquired data according to the corresponding coding, perform topic modeling on the acquired data after the data pre-processing, and obtain the target information data after topic modeling; S3: Information collection of concerned user information: confirm the concerned user information in the target information data, extract user portrait features based on the concerned user information, and mark the extracted user screen features as standard user data; S4: User information feature analysis: Overlap the standard user data and target information data, and determine the proportion of the target information data’s attention group based on the overlapping area; The data cleaning process includes: Extracting text data information from the information data; Performing text recognition on the text data information to determine whether there is a garbled field in the text data information; When there is a garbled field in the text data information, a weight parameter corresponding to the garbled field is obtained according to the parameter information of the garbled field; wherein the weight parameter is obtained by the following formula: Where W represents the weight parameter corresponding to the garbled field; F m Indicates the number of characters corresponding to the garbled field; F j Indicates the number of characters corresponding to the short sentence in which the garbled field is located; F d represents the number of characters corresponding to the paragraph in the paragraph where the garbled field is located; λ represents the cosine similarity between the average word vector corresponding to the upper part of the sentence information of the garbled field and the average word vector corresponding to the lower part of the sentence information; When the weight parameter of the garbled code field is lower than a preset weight threshold, it is determined whether the garbled code field needs to be deleted; When the weight parameter of the garbled code field is not lower than the preset weight threshold, the original data information of the text data information corresponding to the garbled code part is retrieved from the data interface end; When the weight parameter of the garbled code field is lower than a preset weight threshold, determining whether the garbled code field needs to be deleted includes: When the weight parameter of the garbled code field is lower than a preset weight threshold, the upper sentence information and the lower sentence information of the text part where the garbled code field is located are retrieved; Obtaining the Euclidean distance between the average word vector corresponding to the upper part of the sentence information and the average word vector corresponding to the lower part of the sentence information; The weight parameter of the garbled field is compensated by using the Euclidean distance between the average word vector corresponding to the upper part of the sentence information and the average word vector corresponding to the lower part of the sentence information to obtain the weight parameter after compensation; wherein the weight parameter after compensation is obtained by the following formula: Among them, W b represents the weight parameter after compensation processing; O represents the Euclidean distance between the average word vector corresponding to the upper part of the sentence information and the average word vector corresponding to the lower part of the sentence information; O0 represents the preset Euclidean distance threshold; When the weight parameter after the compensation process is lower than the preset weight threshold, the garbled field is deleted; When the weight parameter after the compensation process is not lower than the preset weight threshold, the original data information of the text data information corresponding to the garbled part is retrieved from the data interface end; User portrait feature extraction based on user information in S3, including: Retrieve the concerned browsing users in the target information data; The browsing time data of the followed browsing users are confirmed, and the followed browsing users whose browsing time is within a preset range are taken as target followed browsing users; Retrieve the browsing IP address of the target browsing user; Confirm the target user profile based on the browsing IP address; Among them, user portrait features include the region, age, and gender of the target browsing user; And mark the confirmed user portrait features as standard user data.

2. The method for discovering network information topics and their concerned user groups according to claim 1, characterized in that: The network information data in S1 is obtained through different acquisition platforms, including: The platforms for obtaining online information data include social media platforms, website optimization platforms, third-party data platforms, and manual retrieval platforms; Match the information data obtained from different assessments with the platform respectively; After the correspondence is completed, the acquired information data and the unique coding number of the acquisition platform are obtained.

3. The method for discovering network information topics and their concerned user groups according to claim 2, characterized in that: Data preprocessing is performed on the data to be acquired in S2, including: Confirm the information data in the unique coding number; After confirmation, the information data will be cleaned, normalized and converted in sequence; Among them, data cleaning first processes missing values, abnormal values ​​and duplicate values ​​of information data, and obtains first processed data after processing; Then scaling the first processed data to within a preset range, completing data normalization processing after scaling, and obtaining second processed data; The second processed data is converted into a modeling format, and after the conversion, the data conversion process is completed to obtain the third processed data.

4. The method for discovering network information topics and their concerned user groups according to claim 3, characterized in that: Perform topic modeling on the data to be obtained in S2, including: Before subject modeling is performed on the third processed data, data subject features are first extracted from the third processed data; Among them, first confirm the data subject type of the third processed data, which includes political information subject, economic information subject, cultural information subject and social information subject; Performing data theme analysis on the third processed data, dividing the third processed data into a multi-layer theme tree, each layer including a plurality of clusters; Cluster the sub-data information corresponding to each layer to obtain several cluster sets, and assign each cluster set to the corresponding cluster; Performing a word segmentation operation on the sentence type of the third processed data included in the cluster set in each cluster to obtain a plurality of extracted words in the third processed data; Determine the word feature of each extracted word and the sentence feature of each extracted word in the sentence type.

5. The method for discovering network information topics and their concerned user groups according to claim 4, characterized in that: Topic modeling for the data to be obtained in S2 also includes: Determine the similarity between the extracted words based on the constructed synonym dictionary and the word features of each extracted word; Obtain a word set of extracted words according to the similarity percentage; The extracted words with consecutive positions in the word set are concatenated with the maximum length in the smallest text unit to obtain concatenated words; Clean the concatenated words according to the statistical features of parts of speech to obtain a keyword set; Convert keywords in the keyword set into word vectors; Confirm the distance between each word vector in any cluster and the standard word vector corresponding to each cluster, and select the keyword corresponding to the word vector with the smallest distance as the target keyword in any cluster; The keywords included in each layer of the topic tree are determined according to the target keywords in any cluster.

6. The method for discovering network information topics and their concerned user groups according to claim 5, characterized in that: Topic modeling for the data to be obtained in S2 also includes: Match the corresponding information topic samples in the data topic type according to the keywords; Determine the data classification identifier based on the information subject sample, and at the same time, input the data classification identifier into a preset neural network for learning to determine the classification expression of the data classification identifier; An information topic classification model is constructed according to the classification expression, and at the same time, the information topic samples are input into the information topic classification model for classification; After the classification is completed, the subject type of the third processed data and the modeling model of the subject type are obtained; The topic type of the third processed data and the modeling model of the topic type are labeled as target information data.

7. The method for discovering network information topics and their concerned user groups according to claim 6, characterized in that: The proportion of the concerned groups for the target information data determined based on the overlapping areas in S4 includes: Convert standard user data and target information data into curve data respectively; After the curve data conversion is completed, the curve user data and curve information data are obtained; Then, the curve user data is used to construct sub-curve data according to the user portrait characteristics; Compare the curve information data with the sub-curve data for curve overlap; After the curves overlap, the proportion of the concerned groups of the curve information data is determined according to the overlap rate of the overlap area; The higher the overlap degree between the curve information data and the sub-curve data, the higher the overlap rate between the curve information data and the sub-curve data, and the closer the focus group of the curve information data is to the sub-curve data.

Images