A network emotion data retrieval method and system, an electronic device, and a storage medium

Abstract

The application discloses a network emotion data retrieval method and system, electronic equipment and a storage medium, and belongs to the technical field of internet network emotion interest point data discovery processing. The network emotion data retrieval method comprises the following steps: splitting an interest point expression into a plurality of keywords according to a logical symbol, and setting an interest point expression in which a keyword matched with to-be-matched stream content is located as a candidate interest point expression; logically disassembling the candidate interest point expression to obtain a recursive encapsulation data structure; and matching the to-be-matched stream content with the recursive encapsulation data structure to obtain a network emotion data retrieval result. The application can improve the accuracy of retrieving network emotion data while ensuring the recall rate.

Description

Technical Field

[0001] This application relates to the field of Internet interest point data discovery and processing, and in particular to a method, system, electronic device and storage medium for retrieving Internet interest data. Background Technology

[0002] Internet sentiment data forms the information foundation for building a cybersecurity system. Related technologies typically rely on real-time matching of real-time data content using keyword interest expressions to discover points of interest. However, this approach depends on users' personalized, subjective keyword configurations. When users configure a large number of scenario-based keywords, it can exceed the search engine's MaxBooleanClauses query limit, causing parsing anomalies and impacting search engine stability. Furthermore, it doesn't consider the part-of-speech tagging of keywords related to region and people's names, leading to insufficient data retrieval and false positives.

[0003] Therefore, how to improve the accuracy of retrieving online sentiment data while ensuring recall is a technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0004] The purpose of this application is to provide a method for retrieving internet information data, a system for retrieving internet information data, an electronic device, and a storage medium that can improve the accuracy of retrieving internet information data while ensuring recall.

[0005] To address the aforementioned technical problems, this application provides a method for retrieving internet traffic data, which includes:

[0006] The interest point expression is split into multiple keywords according to logical symbols, and the interest point expression containing the keyword that matches the content of the stream to be matched is set as the candidate interest point expression;

[0007] The candidate interest point expressions are logically decomposed to obtain a recursive encapsulated data structure;

[0008] The content of the stream to be matched is matched with the recursive encapsulated data structure to obtain the network information data retrieval results.

[0009] Optionally, the interest point expression containing the keyword that matches the content of the stream to be matched can be set as an alternative interest point expression, including:

[0010] A keyword dictionary is constructed based on parts of speech; wherein, the elements in the keyword dictionary are an inverted set mapping relationship from the part-of-speech keyword to the interest point expression ID;

[0011] The content to be matched is matched with the keywords in the keyword dictionary, and the interest point expression of the matched keyword is set as the alternative interest point expression;

[0012] Correspondingly, it also includes:

[0013] Add the candidate point of interest expressions to the pre-match set.

[0014] Optionally, the candidate interest point expressions are logically decomposed to obtain a recursive encapsulated data structure, including:

[0015] The candidate interest point expressions are logically decomposed to obtain a recursive encapsulated data structure corresponding to each candidate interest point expression, and the recursive encapsulated data structure is serialized.

[0016] The recursive encapsulation data structure includes an ANDS array, an ORS array, finest-grained keywords, and keyword part-of-speech tags. The ANDS array is a list of expressions representing logical AND operations in the candidate point of interest expressions, and the ORS array is a list of expressions representing logical OR operations in the candidate point of interest expressions.

[0017] Optionally, the content of the stream to be matched is matched with the recursively encapsulated data structure to obtain the network information retrieval results, including:

[0018] The finest-grained keywords are used to match the content of the stream to be matched, and a first matching result is obtained;

[0019] And / or, use the ANDS array to match the content of the stream to be matched to obtain a second matching result;

[0020] And / or, use the ORS array to match the content of the stream to be matched to obtain a third matching result;

[0021] Determine whether the first matching result, the second matching result, or the third matching result is a successful match;

[0022] If so, then the content of the stream to be matched will be set as the result of the internet data retrieval.

[0023] If not, the online data retrieval is deemed unsuccessful.

[0024] Optionally, after obtaining the internet sentiment data retrieval results, the following may also be included:

[0025] Generate a text DAG graph corresponding to the network data retrieval results;

[0026] The maximum probability path is calculated using the text DAG graph to obtain the word segmentation results;

[0027] The search results for internet data are filtered according to the word segmentation results in order to remove text that does not conform to the word segmentation results.

[0028] Optionally, after obtaining the internet sentiment data retrieval results, the following may also be included:

[0029] The pre-trained annotation model is used to perform named entity recognition on the internet sentiment data retrieval results. The internet sentiment data retrieval results are then filtered based on the named entity recognition results to remove text that does not conform to the named entity recognition results.

[0030] Optionally, after obtaining the internet sentiment data retrieval results, the following may also be included:

[0031] The search results for the online sentiment data are filtered using a hierarchical knowledge graph of regional term inference to remove text that does not conform to the hierarchical knowledge graph of regional term inference.

[0032] This application also provides a system for retrieving internet information data, the system comprising:

[0033] The expression filtering module is used to split the point of interest expression into multiple keywords according to logical symbols, and set the point of interest expression containing the keyword that matches the content to be matched as the candidate point of interest expression;

[0034] The logic decomposition module is used to logically decompose the candidate point of interest expression to obtain a recursive encapsulated data structure.

[0035] The matching module is used to match the content of the stream to be matched with the recursive encapsulated data structure to obtain the network information data retrieval results.

[0036] This application also provides a storage medium on which a computer program is stored, wherein the computer program, when executed, implements the steps of the above-described method for retrieving network information data.

[0037] This application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor invokes the computer program in the memory to implement the steps of the above-described method for retrieving online information data.

[0038] This application provides a method for retrieving internet sentiment data, comprising: splitting an interest point expression into multiple keywords according to logical symbols; setting the interest point expression containing the keyword that matches the content to be matched as a candidate interest point expression; logically decomposing the candidate interest point expression to obtain a recursive encapsulated data structure; and matching the content to be matched with the recursive encapsulated data structure to obtain internet sentiment data retrieval results.

[0039] This application decomposes point-of-interest (POI) expressions into multiple keywords and determines candidate POI expressions based on the matching results between the content to be matched and the keywords. This application logically decomposes the candidate POI expressions to obtain a recursively encapsulated data structure, and uses this recursively encapsulated data structure to retrieve content from the content to be matched that matches the candidate POI expressions, thereby obtaining the internet information data retrieval results. This application utilizes a recursively encapsulated data structure for retrieval, effectively avoiding the impact of complex, long-nested expression queries on the stability of the indexing system. Therefore, this application can improve the accuracy of internet information data retrieval while maintaining recall. This application also provides an internet information data retrieval system, a storage medium, and an electronic device, all with the aforementioned beneficial effects, which will not be elaborated upon here. Attached Figure Description

[0040] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 A flowchart illustrating a method for retrieving internet information data provided in this application embodiment;

[0042] Figure 2 This is a schematic diagram of the recursive matching tree corresponding to the encapsulated data structure provided in the embodiments of this application;

[0043] Figure 3 A DAG probability graph of the stream content to be matched is provided in an embodiment of this application;

[0044] Figure 4 This application provides a hierarchical knowledge graph for regional terminology reasoning.

[0045] Figure 5 This is a schematic diagram of the structure of a network information retrieval system provided in an embodiment of this application. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0047] Please see below. Figure 1 , Figure 1This is a flowchart illustrating a method for retrieving internet information data provided in an embodiment of this application.

[0048] Specific steps may include:

[0049] S101: Split the point of interest expression into multiple keywords according to logical symbols, and set the point of interest expression containing the keyword that matches the content of the stream to be matched as the candidate point of interest expression;

[0050] This embodiment can be applied to a computer or server with data retrieval function. Before this step, it can receive the retrieval command input by the user and determine the point of interest expression and retrieval object (the content to be matched stream) to be used based on the retrieval command.

[0051] Interest expressions refer to a set of user-defined complex logical keyword logical expressions configured in the online sentiment monitoring system. These expressions support operations using logical symbols such as +, |, -, (, ), representing AND, OR, NOT, and logical closure, respectively. For example, if a user is interested in negative online sentiment information related to the catering industry in the Development Zone of City A, the logical expression 2-1 would be configured, such as: (Development Zone of City A | Development Zone) + (Food Safety | Closure | Exposure | Complaint | Bankruptcy | Debt | Recruitment | Supply Chain Transfer | Compensation | Bankruptcy | Closure). For even greater specificity, keywords for specific sub-regions and targeted monitoring expressions for designated units or enterprises can be added.

[0052] This step allows you to break down the interest expression into multiple keywords based on the position of the logical symbols within it. Taking the expression above as an example, the resulting keywords include: City A Development Zone, Development Zone, Food Safety, Business Closure, Exposure, Complaint, Bankruptcy, Debt, Recruitment, Supply Chain Relocation, Compensation, Bankruptcy, Closure.

[0053] In this embodiment, there can be multiple interest expressions. After obtaining the keywords corresponding to the interest expression, this application matches the content to be matched with the keywords, and then sets the interest expression containing the matched keywords as the candidate interest expression.

[0054] S102: Logically decompose the expression of the candidate point of interest to obtain a recursive encapsulated data structure;

[0055] In this step, after obtaining the candidate interest point expressions, the expressions are logically decomposed to construct a recursive encapsulated data structure for each candidate expression. After obtaining the recursive encapsulated data structure, it can be cached and serialized to prevent unnecessary duplicate construction.

[0056] S103: Match the content of the stream to be matched with the recursive encapsulated data structure to obtain the network information data retrieval result.

[0057] Specifically, this application can match the content of the stream to be matched with the recursive encapsulation data structure corresponding to each candidate point of interest expression, thereby determining the text content in the stream to be matched that conforms to the recursive encapsulation data structure, and using the above text content as the result of internet sentiment data retrieval.

[0058] This embodiment decomposes the point of interest expression into multiple keywords and determines candidate point of interest expressions based on the content to be matched and the keywords. This embodiment logically decomposes the candidate point of interest expressions to obtain a recursive encapsulated data structure. The recursive encapsulated data structure is then used to retrieve content from the content to be matched that matches the candidate point of interest expressions, thereby obtaining the internet data retrieval results. This embodiment utilizes a recursive encapsulated data structure for retrieval, effectively avoiding the impact of complex, long nested expression queries on the stability of the indexing system. Therefore, this embodiment can improve the accuracy of internet data retrieval while maintaining recall.

[0059] As for Figure 1 A further description of the corresponding embodiment suggests that candidate point of interest expressions can be set in the following way: A keyword dictionary is constructed based on parts of speech; wherein, the elements in the keyword dictionary are inverted sets mapping relationships from keywords of the corresponding parts of speech to point of interest expression IDs; the content to be matched is matched against the keywords in the keyword dictionary, and the point of interest expression containing the matched keyword is set as the candidate point of interest expression. Furthermore, after obtaining the candidate point of interest expressions, all candidate point of interest expressions can be added to a pre-matching set for logical decomposition of the candidate point of interest expressions in the pre-matching set.

[0060] As for Figure 1 A further description of the corresponding embodiment reveals that the recursive encapsulated data structure can be obtained through logical decomposition as follows: The candidate interest point expressions are logically decomposed to obtain a recursive encapsulated data structure corresponding to each candidate interest point expression, and the recursive encapsulated data structure is then serialized. The recursive encapsulated data structure includes an ANDS array, an ORS array, finest-grained keywords, and keyword parts of speech. The ANDS array is a list of expressions representing logical AND operations in the candidate interest point expressions, and the ORS array is a list of expressions representing logical OR operations in the candidate interest point expressions.

[0061] Furthermore, this embodiment can utilize a recursive encapsulated data structure to perform the following matching operations to obtain internet traffic data retrieval results: matching the content to be matched using the finest-grained keywords to obtain a first matching result; and / or, matching the content to be matched using the ANDS array to obtain a second matching result; and / or, matching the content to be matched using the ORS array to obtain a third matching result; determining whether the first matching result, the second matching result, or the third matching result is a successful match; if yes, setting the content to be matched as the internet traffic data retrieval result; if no, determining that the internet traffic data retrieval has failed. In the above process, as long as a match is successful using the ANDS array, the ORS array, or the finest-grained keywords, the successfully matched text content in the content to be matched can be set as the internet traffic data retrieval result.

[0062] As for Figure 1 Further, in accordance with the embodiments, after obtaining the internet sentiment data retrieval results, keyword segmentation and part-of-speech tagging can be performed based on statistical natural language models and deep learning models to reduce false recalls. The specific process includes: generating a text DAG (Directed Acyclic Graph) corresponding to the internet sentiment data retrieval results; calculating the maximum probability path using the text DAG to obtain word segmentation results; and filtering the internet sentiment data retrieval results according to the word segmentation results to remove text that does not conform to the word segmentation results. Furthermore, after obtaining the internet sentiment data retrieval results, a pre-trained annotation model can be used to perform named entity recognition on the internet sentiment data retrieval results, and the results can be filtered based on the named entity recognition results to remove text that does not conform to the named entity recognition results. The aforementioned pre-trained annotation model can be a pre-trained annotation model based on a deep learning bidirectional GRU (Recurrent Neural Network) + CRF (Conditional Random Field). GRU is a variant of the LSTM (Laser-Swept Memory Model) and was proposed to address issues such as long-term memory and gradients in backpropagation. Compared to LSTM, GRU is simpler to compute and easier to implement.

[0063] As for Figure 1 In a further description of the corresponding embodiment, after obtaining the internet sentiment data retrieval results, effective recall can also be achieved based on the reasoning mechanism of regional keywords. The specific process is as follows: the internet sentiment data retrieval results are filtered using a regional word reasoning hierarchy knowledge graph in order to remove texts that do not conform to the regional word reasoning hierarchy knowledge graph.

[0064] The process described in the above embodiments is illustrated below through examples in practical applications.

[0065] This embodiment provides an optimization scheme for matching network sentiment interest points based on complex expressions. This scheme can optimize key issues in network sentiment data matching, improve the recall and accuracy of interest point data matching, and enhance user satisfaction.

[0066] Currently, there are two common solutions for retrieving online sentiment data: Solution 1: Based on data segmentation and indexing, the search engine query parser performs query parsing and matching based on user-configured interest expressions, relying on index matching based on logical expressions of pure keywords; Solution 2: Based on the analysis of massive user behavior click data by big data analysis, user profile tags are constructed by extracting user browsing and click data content, and models trained on click data are used to predict potentially matching interest data.

[0067] The first approach, based entirely on keyword interest expressions, uses real-time matching of real-time data to discover interest points. However, due to its reliance on user-defined keywords, a large number of user-specified scenario terms can exceed the search engine's MaxBooleanClauses query limits, causing parsing errors and impacting search engine stability. Furthermore, it fails to consider part-of-speech tagging for keywords related to region and people, leading to insufficient data recall and false positives. The second approach, based on user behavior log analysis, requires training a model using user browsing data to build user profiles. This approach is suitable for 2C businesses, such as short videos and news aggregators, which generate massive amounts of behavioral data and target a large user base. However, it is unsuitable for the 2B online sentiment monitoring and early warning system addressed in this system. Because online sentiment monitoring systems have high requirements for recall, missing matching of customer-specified keywords can lead to serious user complaints; therefore, a recommendation scheme based on user behavior is unsuitable.

[0068] To meet the requirement of improving matching performance while ensuring recall in online sentiment analysis systems, this embodiment proposes a streaming content memory matching scheme based on part-of-speech tagging based on interest expressions. It then enriches the interest expressions by expanding them with a regional dictionary. Based on these optimizations, the logical expressions of interest points are improved to ensure data recall and matching accuracy. This embodiment includes the following three parts:

[0069] The first part, the memory matching process based on keyword-to-interest inverted index and logical encapsulation, is as follows:

[0070] First, let's look at two examples of logical expressions for keywords representing user interests in the configuration:

[0071] Configuration-1: Enterprise - Health Monitoring

[0072] Point of Interest Expression 1: (City A|Fenhu|Shengze|Tongli)#LOC+(Company E)#ORG+(Default|Dishonesty|Bankruptcy|Credit|Debt|Layoffs|Compensation|Closed-off|Shutdown|Penalty|Resumption of Work and Production|Start of Work|Closed-off|Closed-off|Orders|Fraud|Energy Saving|Low Carbon)#Scene Word;

[0073] Configuration-2: Enterprise - Environmental Monitoring

[0074] Point of Interest Expression 2: (City A)#LOC+(Energy Saving|Low Carbon|Recycling|Sewage|Waste Gas|Pollution|Exposure|Complaint|Report)#Scene Word;

[0075] In the above expressions, LOC represents a geographical term, and ORG identifies a person or organization.

[0076] The content to be matched is: "#F District E Company Fraud# E Company has engaged in fraudulent consumer behavior. Its sales staff member, Xia**, refused to sell me the goods after signing the contract. The contract clearly states xx. On the grounds that E Company was unwilling to sell to me at the agreed discount, I was forced to return the goods. I have filed a complaint during this period."

[0077] The specific steps of the memory matching processing technology based on expression logic encapsulation provided in this embodiment are as follows:

[0078] Step 1: Segment the interest point expression into keywords according to logical symbols. This step does not consider logical relationships. Construct three keyword dictionaries based on the part of speech of the expression, where each element is an inverted set mapping relationship from the keyword of its part of speech to the interest point ID:

[0079] ① Dictionary of regional terms: {City A->{1,2}, Fenhu->{1}, Shengze->{1}, Tongli->{1}}

[0080] ② Dictionary of people and organizations: {Company E -> {1}}

[0081] ③ Event Scene Word Dictionary: {Breach of Contract->{1}, Dishonesty->{1}, Bankruptcy->{1}, Credit->{1}, Debt->{1}, Layoffs->{1}, Compensation->{1}, Closure->{1}, Shutdown->{1}, Punishment->{1}, Resumption of Work and Production->{1}, Start of Work->{1}, Closure->{1}, Bankruptcy->{1}, Orders->{1}, Fraud->{1}, Energy Saving->{1,2}, Low Carbon->{1,2}, Recycling->{2}, Sewage->{2}, Waste Gas->{2}, Pollution->{2}, Exposure->{2}, Complaint->{2}, Report->{2}}.

[0082] The keyword inverted index and matching encapsulation query calculation techniques based on in-memory batch data effectively avoid the impact of complex long nested expression queries on the query stability of the index system, and are easier to scale in a distributed manner.

[0083] Step 2: For the content to be matched: if at least one keyword in the three keyword dictionaries matches, then the point of interest expression ID is added to the set. Based on this logic, point of interest expression 1 and point of interest expression 2 can be added to the pre-matching set S according to the keyword "City A". pre .

[0084] Step 3: Combine the pre-matched set of interest points S pre Interest point expressions 1 and 2 are logically decomposed, and each interest point expression constructs a recursive encapsulated data structure KeyWords. The recursive encapsulated data structure KeyWords contains an ANDS array, an ORS array attribute, the finest-grained keyword ele, and the corresponding part-of-speech tag posTag. The ANDS array is a list of expressions that must be matched together, and the ORS array is a list of expressions that match if at least one expression is satisfied. The structure is shown in Table 1.

[0085] Table 1. Encapsulated Data Structures and Basic Matching Algorithms

[0086]

[0087]

[0088] During the matching process between the stream content to be matched and the encapsulated data structure, a fine-grained keyword non-empty check is first performed. If the keyword is not empty, a matching check is performed and the result is returned. Otherwise, an ORS recursive matching check is performed. If any item in the ORS list is satisfied, the point of interest is matched; otherwise, an ANDS check is performed. If the subsets are jointly satisfied, the point of interest is matched; otherwise, the stream content does not match the point of interest. Please refer to [link to relevant documentation]. Figure 2 , Figure 2 This is a schematic diagram of the recursive matching tree corresponding to the encapsulated data structure provided in this application embodiment. After recursively constructing the parse tree and recursively matching the stream content in step 3, the matching of interest point 2 can be excluded because the stream content does not include the LOC keyword "Tongli"; the matching of interest point 1 is retained because it simultaneously satisfies the keywords "Tongli, fraud, E company". At the same time, the recursive expression data structure KeyWords constructed in this step can be cached and serialized to prevent the computational overhead of repeated construction.

[0089] The above process encapsulates and performs matching calculations based on in-memory expression data structures, constructing an inverted document ID index from keywords to points of interest for pre-matching a set S. preThe calculation, and then for S that may match points of interest. pre The calculation is based on an expression decomposition and matching scheme. This embodiment can be combined with a distributed computing framework to perform matching operations on streaming content, solving the problem of rapid matching of massive amounts of data and numerous points of interest, and reducing reliance on search engine queries.

[0090] The second part, the expression reasoning mechanism based on statistics, part-of-speech detection, and regional dictionary expansion, is as follows:

[0091] This embodiment retains S in the first part. pre Interest point 1 was identified, but manual observation revealed that the matched word "Tongli" had a context related to "contractually clear," indicating it was not a Tongli-related word in our regional part-of-speech tagging set. Therefore, the match for interest point expression 1 should be temporarily excluded. Thus, an automated method is needed for this determination. This embodiment proposes a sentence segmentation method based on maximum probability prediction and a keyword detection method based on pre-trained models for part-of-speech tagging to reduce false positives.

[0092] The reasoning process based on statistics and part-of-speech tagging is as follows: Using a maximum probability algorithm based on a statistical language model and dynamic programming, it determines whether a matching word can be formed. The basic logic is as follows: Figure 3 As shown. Figure 3 This application provides a DAG probability graph of the stream content to be matched. Figure 3 In this example, S represents the start, E represents the end, and 0, 1, 2, 3, and 4 represent the word numbers. This embodiment refines the reasoning to the sentence level; sentences without matching interest point keywords are not segmented, reducing unnecessary computation. The segmentation method is illustrated using the sentence "Contract clearly states" from the example stream content in the first part of this embodiment. Based on the basic dictionary, all DAG paths of the sentence can be obtained, such as... Figure 3 As shown: {0:[0,1],1:[1,2],2:[2],3:[3,4],4:[4]}, therefore there are four ways to divide the contract: "Contract / in / clear / details|Contract / in / clear / details|Contract / in / clear / details". Finally, the maximum probability path is calculated based on the algorithm. The process of calculating the maximum probability path based on the dynamic programming algorithm is as follows:

[0093]

[0094] In the above process, the log calculation of probability product is transformed into probability addition. The method FREQ.get can obtain the word frequency of a segmentation word in the basic main word list. Through the dynamic planning calculation in Table 1, it is found that the maximum probability segmentation path is "contract / in / clear", so the segmentation probability of "contract" is greater than that of "tongli". Based on this segmentation result, the retrieval results of the network sentiment data can be screened to remove the texts that do not conform to the segmentation result.

[0095] This embodiment can also perform named entity recognition based on the popular pre-trained annotation model of deep learning bidirectional GRU+CRF, and analyze that "contract / work category concept" belongs to noun vocabulary and cannot form the target word "tongli / LOC". Therefore, based on this inference, the word "tongli" cannot be matched, so the matching of interest point 1 is also excluded. Through this reasoning mechanism, the following similar ambiguous matching problems can be effectively avoided: "Today the monitor and the mountain village residents inspected xx together", and the interest point geographical words are "heshan, changheshan". Similarly, the similar convection content is: "XX appears on Wujiang Road, Shanghai", and the interest point is: the matching of "Wujiang"#LOC. Through the natural language processing model technology, it is recognized that "Wujiang Road, Shanghai" is a place type word and does not match "Wujiang"#LOC, so it cannot be determined to recall. By calculating the maximum possible segmentation probability of the sentence where the matching word is located and using the trained词性标注模型 for词性检测 of the待匹配词, the recall of non-interest point data can be effectively avoided. In specific implementation, it is not limited to the词性判定模型 adopted in this patent, and multiple advanced deep learning models in the industry can be used for词性校验 and named entity recognition technology for词性校验 and voting decision inference.

[0096] The third part, the process of screening the retrieval results of network sentiment data based on the expansion inference of the geographical dictionary is as follows:

[0097] The geographical attributes of network sentiment data are quite important. The geographical word matching based entirely on segmentation matching is very likely to cause mis-matching or non-recall problems. This patent separately processes the LOC词性 geographical words in the user interest points, and based on the词性检测 inference, expands and matches based on the geographical word expansion. Based on the geographical knowledge graph or the geographical knowledge base of the encyclopedia map, the upper and lower level relationships of geographical knowledge can be obtained. The hierarchical geographical knowledge graph is as Figure 4 shown, Figure 4 is a hierarchical knowledge graph of geographical word reasoning provided by an embodiment of the present application. The figure shows that City A includes Gusu District, Industrial Park, High-tech Zone, Huqiu District, Wuzhong District, Wujiang District and Xiangcheng District; the High-tech Zone and Huqiu District include Fengqiao Street and Shishan Street, and Wujiang District includes Tongli Town, Fenhu Town, and Shengze Town; City B also includes an Industrial Park and a High-tech Zone. The High-tech Zone and Huqiu District refer to the same district, that is, synonymous geographical words.

[0098] Currently, geographic information can be obtained from online sentiment data through various means. Geographic attributes can be extracted using named entity recognition technology derived from natural language processing. Social media platforms have already enabled the display of commenters' geographic locations, WeChat public accounts can also obtain their geographic information, and local forum data on public opinion can also have its geographic attributes obtained through information such as webpage hierarchy and tags.

[0099] As illustrated by the example of the stream content to be matched, the geographical attribute of "Xiangcheng District" was identified through technical means. Based on the geographical knowledge graph, the query found that Xiangcheng District uniquely and unambiguously belongs to Suzhou City in the geographical graph dictionary. For ambiguous geographical areas, a reasoning mechanism based on the co-occurrence of superior, subordinate, and peer geographical terms is needed. Therefore, although the stream content did not match the Suzhou geographical keywords of interest point expression 1 and interest point expression 2, through the hierarchical geographical affiliation reasoning mechanism, the content matched Xiangcheng District, thus satisfying the requirement of uniquely matching the superior geographical area of ​​Suzhou City. Therefore, the example stream content can match interest point expression 1 and interest point expression 2, thereby improving the recall of effective interest data.

[0100] In summary, after in-memory expression matching, natural language processing techniques are used to segment the text of sentences containing keywords to determine the most probable combinations and to perform part-of-speech tagging of target words. Finally, inference matching is performed based on an expanded regional knowledge base. Specifically, in the implementation of the matching system, data based on inference filtering and inference recall can be tagged accordingly, allowing users to individually filter the inference-matched and filtered data during display, thus improving the overall user experience of the online intelligence system.

[0101] The above embodiment proposes a memory-based logical expression matching algorithm, along with segmentation-based word re-checking analysis and a regional word inference mechanism, effectively improving the system's availability and data recall performance. This embodiment can solve at least the following problems: (1) the problem of search engine query parsing errors and inability to respond due to overly complex expressions; (2) the problem of false recall caused by not considering the configuration of keyword part-of-speech detection; (3) the problem of low recall due to not utilizing data regional tag information and the hierarchical relationship of regional dictionaries. Compared with existing solutions, this embodiment significantly improves the matching effect of network interest points based on expression matching.

[0102] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of a network information retrieval system provided in an embodiment of this application;

[0103] The system may include:

[0104] The expression filtering module 501 is used to split the interest point expression into multiple keywords according to logical symbols, and set the interest point expression containing the keyword that matches the content to be matched as the candidate interest point expression.

[0105] The logic decomposition module 502 is used to logically decompose the candidate interest point expression to obtain a recursive encapsulated data structure.

[0106] The matching module 503 is used to match the content of the stream to be matched with the recursive encapsulated data structure to obtain the network information data retrieval result.

[0107] This embodiment decomposes the point of interest expression into multiple keywords and determines candidate point of interest expressions based on the matching results between the content to be matched and the keywords. This embodiment logically decomposes the candidate point of interest expressions to obtain a recursive encapsulated data structure. The recursive encapsulated data structure is then used to retrieve content from the content to be matched that matches the candidate point of interest expressions, thereby obtaining the internet data retrieval results. This embodiment utilizes a recursive encapsulated data structure for retrieval, effectively avoiding the impact of complex, long-nested expression queries on the stability of the indexing system. Therefore, this embodiment can improve the accuracy of internet data retrieval while maintaining recall.

[0108] Furthermore, the process by which the expression filtering module 501 sets the interest point expressions containing keywords that match the content to be matched as candidate interest point expressions includes: constructing a keyword dictionary based on parts of speech; wherein, the elements in the keyword dictionary are inverted sets mapping relationships from keywords of the corresponding parts of speech to interest point expression IDs; matching the content to be matched with the keywords in the keyword dictionary, and setting the interest point expressions containing the matched keywords as candidate interest point expressions; correspondingly, the expression filtering module 501 is also used to add the candidate interest point expressions to the pre-matching set.

[0109] Furthermore, the logic decomposition module 502 is used to logically decompose the candidate interest point expressions to obtain a recursive encapsulated data structure corresponding to each candidate interest point expression, and to serialize the recursive encapsulated data structure.

[0110] The recursive encapsulation data structure includes an ANDS array, an ORS array, finest-grained keywords, and keyword part-of-speech tags. The ANDS array is a list of expressions representing logical AND operations in the candidate point of interest expressions, and the ORS array is a list of expressions representing logical OR operations in the candidate point of interest expressions.

[0111] Furthermore, the matching module 503 matches the content to be matched with the recursive encapsulated data structure to obtain the network situation data retrieval result, including: matching the content to be matched with the finest-grained keywords to obtain a first matching result; and / or, matching the content to be matched with the ANDS array to obtain a second matching result; and / or, matching the content to be matched with the ORS array to obtain a third matching result; determining whether the first matching result, the second matching result, or the third matching result is a successful match; if yes, then setting the content to be matched as the network situation data retrieval result; if no, then determining that the network situation data retrieval has failed.

[0112] Furthermore, it also includes:

[0113] The word segmentation module is used to generate a text DAG graph corresponding to the internet sentiment data retrieval results after obtaining the internet sentiment data retrieval results; it is also used to perform maximum probability path calculation using the text DAG graph to obtain word segmentation results; and it is also used to filter the internet sentiment data retrieval results according to the word segmentation results in order to remove text that does not conform to the word segmentation results.

[0114] Furthermore, it also includes:

[0115] The entity recognition module is used to perform named entity recognition on the internet situation data retrieval results using a pre-trained annotation model after obtaining the internet situation data retrieval results, and to filter the internet situation data retrieval results based on the named entity recognition results in order to remove text that does not conform to the named entity recognition results.

[0116] Furthermore, it also includes:

[0117] The regional reasoning module is used to filter the internet sentiment data retrieval results using a regional term reasoning hierarchical knowledge graph after obtaining the results, so as to remove text that does not conform to the regional term reasoning hierarchical knowledge graph.

[0118] Since the embodiments of the system part correspond to the embodiments of the method part, please refer to the description of the embodiments of the method part for the embodiments of the system part, and they will not be repeated here.

[0119] This application also provides a storage medium on which a computer program is stored, which, when executed, can perform the steps provided in the above embodiments. The storage medium may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0120] This application also provides an electronic device that may include a memory and a processor. The memory stores a computer program, and when the processor calls the computer program in the memory, it can implement the steps provided in the above embodiments. Of course, the electronic device may also include various network interfaces, power supplies, and other components.

[0121] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to in the method section. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0122] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for retrieving online sentiment data, characterized in that, include: The interest point expression is split into multiple keywords according to logical symbols, and the interest point expression containing the keyword that matches the content of the stream to be matched is set as the candidate interest point expression; The candidate interest point expressions are logically decomposed to obtain a recursive encapsulated data structure; The content to be matched is matched with the recursive encapsulated data structure to obtain the network information data retrieval results; The step of setting the interest point expression containing the keyword that matches the content of the stream to be matched as the candidate interest point expression includes: A keyword dictionary is constructed based on parts of speech; wherein, the elements in the keyword dictionary are an inverted set mapping relationship from the part-of-speech keyword to the interest point expression ID; The content to be matched is matched with the keywords in the keyword dictionary, and the interest point expression of the matched keyword is set as the alternative interest point expression; Correspondingly, it also includes: Add the candidate point of interest expressions to the pre-matching set; The candidate point of interest expressions are logically decomposed to obtain a recursive encapsulated data structure, including: The candidate interest point expressions are logically decomposed to obtain a recursive encapsulated data structure corresponding to each candidate interest point expression, and the recursive encapsulated data structure is serialized. The recursive encapsulation data structure includes an ANDS array, an ORS array, finest-grained keywords, and keyword part-of-speech tags. The ANDS array is a list of expressions representing logical AND operations in the candidate interest point expressions, and the ORS array is a list of expressions representing logical OR operations in the candidate interest point expressions. Specifically, matching the content of the stream to be matched with the recursive encapsulated data structure to obtain the network information retrieval results includes: The finest-grained keywords are used to match the content of the stream to be matched, and a first matching result is obtained; The ANDS array is used to match the content of the stream to be matched, and a second matching result is obtained; The ORS array is used to match the content of the stream to be matched, and a third matching result is obtained; Determine whether the first matching result, the second matching result, or the third matching result is a successful match; If so, then the content of the stream to be matched will be set as the result of the internet data retrieval. If not, the online data retrieval is deemed unsuccessful.

2. The method for retrieving internet data according to claim 1, characterized in that, After obtaining the internet sentiment data retrieval results, the following are also included: Generate a text DAG graph corresponding to the network data retrieval results; The maximum probability path is calculated using the text DAG graph to obtain the word segmentation results; The search results for internet data are filtered according to the word segmentation results in order to remove text that does not conform to the word segmentation results.

3. The method for retrieving internet data according to claim 1, characterized in that, After obtaining the internet sentiment data retrieval results, the following are also included: The pre-trained annotation model is used to perform named entity recognition on the internet sentiment data retrieval results. The internet sentiment data retrieval results are then filtered based on the named entity recognition results to remove text that does not conform to the named entity recognition results.

4. The method for retrieving internet data according to claim 1, characterized in that, After obtaining the internet sentiment data retrieval results, the following are also included: The search results for the online sentiment data are filtered using a hierarchical knowledge graph of regional term inference to remove text that does not conform to the hierarchical knowledge graph of regional term inference.

5. A system for retrieving online sentiment data, characterized in that, include: The expression filtering module is used to split the point of interest expression into multiple keywords according to logical symbols, and set the point of interest expression containing the keyword that matches the content to be matched as the candidate point of interest expression; The logic decomposition module is used to logically decompose the candidate point of interest expression to obtain a recursive encapsulated data structure. The matching module is used to match the content of the stream to be matched with the recursive encapsulated data structure to obtain the network information data retrieval results; The process by which the expression filtering module sets the interest point expressions containing keywords that match the content to be matched as candidate interest point expressions includes: constructing a keyword dictionary based on parts of speech; wherein, the elements in the keyword dictionary are an inverted set mapping relationship from the keywords of the part of speech to the interest point expression IDs; matching the content to be matched with the keywords in the keyword dictionary, and setting the interest point expressions containing the matched keywords as the candidate interest point expressions; correspondingly, the expression filtering module is also used to add the candidate interest point expressions to the pre-matching set; The logic decomposition module is used to logically decompose the candidate interest point expressions to obtain a recursive encapsulation data structure corresponding to each candidate interest point expression, and to serialize the recursive encapsulation data structure. The recursive encapsulation data structure includes an ANDS array, an ORS array, finest-grained keywords, and keyword parts of speech. The ANDS array is a list of expressions representing logical AND operations in the candidate interest point expressions, and the ORS array is a list of expressions representing logical OR operations in the candidate interest point expressions. The process by which the matching module matches the content to be matched with the recursive encapsulated data structure to obtain the network situation data retrieval result includes: matching the content to be matched with the finest-grained keywords to obtain a first matching result; matching the content to be matched with the ANDS array to obtain a second matching result; matching the content to be matched with the ORS array to obtain a third matching result; determining whether the first matching result, the second matching result, or the third matching result is a successful match; if yes, then setting the content to be matched as the network situation data retrieval result; if no, then determining that the network situation data retrieval has failed.

6. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor, when calling the computer program in the memory, implements the steps of the method for retrieving network information data as described in any one of claims 1 to 4.

7. A storage medium, characterized in that, The storage medium stores computer-executable instructions, which, when loaded and executed by a processor, implement the steps of the network information data retrieval method as described in any one of claims 1 to 4.

Images