A smart filtering method and system based on target search data

Abstract

This invention relates to the field of intelligent data filtering technology, and more particularly to an intelligent filtering method and system based on target search data. The method includes the following steps: acquiring an original search dataset; preprocessing the original search dataset to generate a standard search dataset; constructing a database based on the standard search dataset to obtain a search database; acquiring user input data; parsing the user input data for keywords to generate user input keyword data; and performing search association on the search database based on the user input keyword data to generate a first search association dataset. This invention improves the accuracy and intelligence of search results through data preprocessing, keyword parsing, attention weight allocation, and intelligent recommendation.

Description

Technical Field

[0001] This invention relates to the field of intelligent data filtering technology, and in particular to an intelligent filtering method and system based on target search data. Background Technology

[0002] With the explosive growth of internet content, researchers have begun exploring smarter filtering methods. Recommender systems have become a significant area of ​​development, leveraging users' historical behavior and interests to recommend relevant content. This approach significantly improves the efficiency and accuracy of information retrieval by analyzing user preferences and behavioral patterns to provide personalized recommendations. However, traditional recommender systems still suffer from limitations such as data sparsity and the cold-start problem. To address these issues, intelligent filtering methods based on target search data have begun to emerge. These methods utilize users' explicit goals or needs, combining big data and machine learning techniques to achieve more accurate data filtering and recommendations. With the development of deep learning and natural language processing technologies, intelligent filtering methods based on target search data have made great strides. Models can understand users' intentions from their target descriptions and filter data based on context and linguistic information. This approach not only improves search accuracy but also handles more complex queries and needs, providing users with more personalized and precise services. However, current traditional search methods cannot effectively address the problem of information overload because they rely primarily on keyword matching without considering the user's actual needs and intentions. Furthermore, they are limited by data sparsity and cold start problems, making it impossible to provide accurate personalized recommendations for users, resulting in low precision and intelligence in filtering. Summary of the Invention

[0003] Therefore, it is necessary to provide an intelligent filtering method and system based on target search data to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, an intelligent filtering method based on target search data is provided, the method comprising the following steps:

[0005] Step S1: Obtain the original search dataset; preprocess the original search dataset to generate a standard search dataset; construct a database based on the standard search dataset to obtain the search database;

[0006] Step S2: Obtain user input data; perform keyword parsing on the user input data to generate user input keyword data; perform search association on the search database based on the user input keyword data to generate a first search association dataset; compare the weight of the first search association data with a preset search association weight threshold; when the weight of the first search association data is greater than or equal to the preset search association weight threshold, sort and display the first search association dataset corresponding to the weight of the first search association data to generate first search association sorted data;

[0007] Step S3: Assign attention weights to the first search-related sorted data to generate attention weights for the first search-related data; compare the attention weights of the first search-related data with the preset standard data attention weight thresholds to obtain high search volume filtering data; calculate the similarity of the high search volume filtering data to obtain high search volume similarity; perform secondary filtering on the high search volume filtering data based on the high search volume similarity to obtain second search-related data;

[0008] Step S4: Utilize blockchain to analyze user search behavior preferences in the second search-related data, generating user search behavior preference data; construct intelligent user behavior profiles based on user search behavior preference data, generating user search behavior profiles; utilize deep neural network algorithms to perform intelligent search recommendations based on user search behavior profiles, generating a recommendation result set to perform intelligent filtering of target search data.

[0009] This invention improves data quality and consistency by preprocessing and standardizing raw search data, thereby enabling more efficient construction of the search database. This contributes to improved search engine performance and accuracy. Keyword parsing and search association allow for the provision of relevant search results based on user input. Sorted display helps users find content of interest more quickly, enhancing the search experience. Attention weight allocation and similarity calculation further filter and refine search results, providing users with more relevant and accurate search recommendations. This increases user satisfaction and reduces search result redundancy. By analyzing user search behavior preferences and building behavioral profiles, the search engine can better understand user interests and needs. Combining deep neural network algorithms for intelligent search recommendations provides personalized search results, increasing user satisfaction and search efficiency. Simultaneously, blockchain technology ensures data security and trustworthiness, further enhancing the reliability of the search engine. Therefore, this invention improves the accuracy and intelligence of search results through data preprocessing, keyword parsing, attention weight allocation, and intelligent recommendation.

[0010] Preferably, step S1 includes the following steps:

[0011] Step S11: Obtain the original search dataset;

[0012] Step S12: Clean the original search dataset to generate the original search clean dataset; filter the original search clean dataset to generate the original search filtered dataset; identify the data format of the original search filtered dataset and standardize the data format of the identified original search filtered dataset to generate the search format dataset.

[0013] Step S13: Classify the search format dataset for duplicate data to obtain non-duplicate search data and duplicate search data, and remove duplicate search data;

[0014] Step S14: Perform missing value detection on the search non-duplicate data to generate missing value detection results for the search non-duplicate data; use the missing value detection results to repair the search non-duplicate data and generate a standard search dataset;

[0015] Step S15: Construct a database based on the standard search dataset to obtain the search database.

[0016] This invention cleanses and filters the original search dataset, removing invalid or redundant data and improving data quality and accuracy. Data format identification and standardization of the original search filtered dataset ensures data consistency and uniformity, facilitating subsequent data processing and analysis. Duplicate classification of the search format dataset identifies and removes duplicate data, avoiding the influence of redundant information and providing a cleaner and more efficient dataset. Missing value detection on non-duplicate search data reveals missing values, allowing for data repair and filling in missing values, making the data more complete and reliable. A search database is built based on the standard search dataset, providing a structured and well-organized database that supports rapid data querying and retrieval, facilitating user search and analysis.

[0017] Preferably, step S14 includes the following steps:

[0018] Step S141: Use a conditional random field to detect missing values ​​in the search non-repeating data and obtain the missing data detection results for the search non-repeating data;

[0019] Step S142: Based on the search results for missing non-duplicate data, impute missing values ​​in the preliminary detected abnormal dataset to generate a search-impacted dataset; perform data consistency checks on the search-impacted data to obtain data consistency check results; and use the data consistency check results to identify data anomalies in the search-impacted dataset to generate data anomaly type identification results.

[0020] Step S143: When the data anomaly type identification result is confirmed to be a semantic anomaly, the knowledge graph method is used to perform semantic repair on the search-filled dataset to generate a semantic repair dataset.

[0021] Step S144: When the data anomaly type identification result is confirmed to be a statistical anomaly, Bayesian inference is used to perform statistical repair on the search-filled dataset to generate a statistically repaired dataset.

[0022] Step S145: Perform repair and verification on the semantic repair dataset and the statistical repair dataset to generate a standard search dataset.

[0023] This invention identifies missing values ​​in the search non-duplicate data by performing missing value detection, which helps to discover dataset incompleteness. Based on the missing value detection results, the initially detected abnormal dataset is imputed to generate a search-imputed dataset. This completes the dataset, making it more comprehensive and improving the accuracy of subsequent analysis. Data consistency is then checked on the search-imputed data to obtain consistency results. This helps identify inconsistencies in the data, such as differences in data format and data type, further improving data quality and consistency. Based on the consistency detection results, data anomalies are identified in step S142, and corresponding repairs are performed. If the anomaly is confirmed as a semantic anomaly (step S143), semantic repair can be performed using knowledge graph methods to improve the accuracy of data semantics. If the anomaly is confirmed as a statistical anomaly (step S144), statistical repair can be performed using Bayesian inference methods to improve the statistical properties of the data. The repaired semantically and statistically repaired datasets are then validated to generate a standard search dataset. This verifies the effectiveness and correctness of the repair and ensures that the generated dataset meets the expected standards and requirements.

[0024] Preferably, step S2 includes the following steps:

[0025] Step S21: Obtain user input data;

[0026] Step S22: Identify the input type of the user input data to obtain user input type data; perform keyword parsing on the user input data based on the user input type data to generate user input keyword data;

[0027] Step S23: Perform search association on the search database based on the user-input keyword data to generate a first search association dataset; calculate the association weights on the first search association dataset to obtain the weights of the first search association data;

[0028] Step S24: Compare the weight of the first search-related data with the preset search-related weight threshold. When the weight of the first search-related data is greater than or equal to the preset search-related weight threshold, sort and display the first search-related dataset corresponding to the weight of the first search-related data to generate the first search-related sorted data.

[0029] This invention, through keyword parsing and association weight calculation, enables the system to more accurately identify user intent, thereby providing more relevant search results. By recognizing input types and parsing specific keywords, the system can effectively process different types of input data, improving data processing efficiency and accuracy. By setting search association weight thresholds, the system can intelligently filter and sort search results, providing users with a more intelligent and personalized search experience. Accurate and efficient search results significantly improve user satisfaction, allowing users to find the information they need more quickly. This process can be applied to different types of data and input methods, giving the system high scalability and adaptability to various application scenarios.

[0030] Preferably, step S22 includes the following steps:

[0031] Step S221: Perform input type recognition on the user input data to obtain user input type data, wherein the input type recognition includes text recognition, speech recognition, image recognition and video recognition;

[0032] Step S222: If the user input data is confirmed to be text, perform text semantic analysis on the user input data to generate text semantic analysis data; extract keywords from the user input data based on the text semantic analysis data to generate text keyword data.

[0033] Step S223: If the user input data is confirmed to be audio, perform a Fourier transform on the user input data to obtain the user input spectrogram; extract the Mel frequency cepstral coefficients from the user input spectrogram to obtain the user input audio features; transcribe the user input audio features into text to generate audio conversion keyword data.

[0034] Step S224: When it is confirmed that the user input type data is an image, image content recognition is performed on the user input data to generate image content recognition data, wherein image content recognition includes image object recognition and image scene recognition; the image content recognition data is used to describe the image content and generate image conversion keyword data.

[0035] Step S225: When it is confirmed that the user input data is video, the video frames are extracted from the user input data based on a preset time interval to obtain the user input video frame set; the frame set change is filtered in the user input video frame set to obtain the user input change frame set; the frame set image content is recognized in the user input change frame set to generate frame set image content data; the frame set image content data is described in the frame set image content data to generate video conversion keyword data;

[0036] Step S226: Perform comprehensive keyword extraction on the text keyword data, audio-to-keyword data, image-to-keyword data, and video-to-keyword data to obtain the user input keyword data.

[0037] This invention enables the system to adapt to diverse user needs and input methods by processing different types of user input (text, audio, images, and video). Through specific processing and keyword extraction for each input type, the system generates more accurate and relevant keyword data, improving search accuracy. The step-by-step processing, optimized for each specific input type, ensures both efficiency and accuracy. Utilizing semantic analysis, spectral analysis, and content recognition technologies, the system deeply understands the content of user input, thereby generating more valuable keywords. Comprehensive processing of different types of keyword data ensures that the final generated keyword data fully reflects the user's search intent, improving the relevance of search results and user satisfaction. By processing multiple user input methods and providing high-quality search results, the system significantly enhances the user experience.

[0038] Preferably, step S3 includes the following steps:

[0039] Step S31: Collect user feedback data from the first search related sorting data to obtain the first search user feedback data, wherein the feedback data collection includes user click frequency collection and user dwell time collection;

[0040] Step S32: Based on the feedback data from the first search user, assign attention weights to the first search-related ranking data to generate attention weights for the first search-related data;

[0041] Step S33: Compare the attention weight of the first search related data with the preset standard data attention weight threshold. When the attention weight of the first search related data is greater than or equal to the preset standard data attention weight threshold, the first search user feedback data is filtered for high search volume to obtain high search volume filtered data.

[0042] Step S34: Calculate the similarity of the high search volume filtering data using the search similarity calculation formula to obtain the high search volume similarity; perform a second filtering on the high search volume filtering data based on the high search volume similarity to obtain the second search related data.

[0043] This invention leverages user feedback data to better understand user preferences and behaviors, thereby optimizing search result ranking and making results more aligned with user needs. Through attention weight allocation and high-volume search filtering, the system dynamically adjusts its search algorithm to ensure the most relevant search results are prioritized. Collecting data such as click frequency and dwell time allows for more precise analysis of user behavior, improving search result accuracy. Through phased filtering and similarity calculations, the system efficiently processes large amounts of search data, ensuring the quality and performance of search results. Optimized search results better meet user expectations, significantly improving the user search experience and satisfaction. By continuously collecting and analyzing user feedback data, the system continuously improves its search algorithm and results, maintaining the search engine's efficiency and relevance.

[0044] Preferably, the search similarity calculation formula in step S34 is as follows:

[0045]

[0046] In the formula, S(Q,D) represents the similarity between query Q and document D, a represents the lower limit of search time, b represents the upper limit of search time, n represents the number of queries, and w i Let λ represent the weight of the i-th selection feature. i Let q represent the attenuation coefficient of the i-th screening feature. i Let d represent the value of the i-th feature of Q. i Let exp(-λ) represent the value of the i-th feature of document D. i (q i -d i ) 2 ) represents the similarity between the i-th features, g(t) represents the contribution adjustment coefficient as it changes with time point t, and σ represents the standard deviation of the normal distribution.

[0047] This invention analyzes and integrates a search similarity calculation formula. The integral term in the formula accumulates similarity values ​​within the interval [a, b], capturing the similarity changes of queries and documents across different feature spaces and time dimensions. Through integration, dynamic or time-varying similarity calculations can be handled, adapting to different application scenarios. (Summation term) Summing over all n features comprehensively considers the contribution of each feature to the similarity score. This ensures that the similarity calculation covers all relevant features, providing a comprehensive similarity assessment. Weight w iThis represents the weight of each feature i, adjusting its importance in similarity calculation. By adjusting the weights, important features can be prioritized, while the influence of less important features can be ignored or reduced, improving the accuracy and relevance of the calculation. The Gaussian function exp(-λ) i (q i -d i ) 2 )) represents the similarity between the i-th features, as q i and d i The difference increases and decreases exponentially. Parameter λ i This controls the degree of influence of feature differences and provides adjustment for the sensitivity of different features. The Gaussian function ensures that similarity is high when feature differences are small and decreases rapidly when differences are large, which is consistent with reality. The weight function g(t) can capture the change of feature similarity over time, adapting to the processing of dynamic data and time series data. Normalization constant. The normalization of similarity calculation results ensures they conform to a normal distribution, facilitating interpretation and comparison. Standardization reduces potential biases during calculation, providing stable similarity results. While conventional search similarity calculation formulas in the art yield the similarity between query Q and document D, the formula provided in this invention allows for a more accurate calculation of their similarity. The formula considers multiple features and their weights, integrating them through a Gaussian function and integrals to provide a comprehensive similarity assessment. The use of normalization constants and Gaussian functions ensures the stability and robustness of the similarity calculation, reducing the impact of noise and outliers. Through the synergistic effect of multiple parameters, accurate, comprehensive, and stable similarity assessments can be provided in high-volume search data processing, adapting to different application scenarios and needs.

[0048] Preferably, step S4 includes the following steps:

[0049] Step S41: Use blockchain to perform distributed and trusted storage of the second search-related data, and generate second search-related storage data;

[0050] Step S42: Analyze user search behavior preferences based on the second search association storage data to generate user search behavior preference data;

[0051] Step S43: Construct an intelligent user behavior profile based on user search behavior preference data to generate a user search behavior profile; use a deep neural network algorithm to construct a personalized prediction model based on the user search behavior profile to generate a personalized intelligent search recommendation model.

[0052] Step S44: Based on the personalized intelligent search recommendation model, perform intelligent search recommendations on the user input data, generate a recommendation result set, and perform intelligent filtering of the target search data.

[0053] This invention utilizes blockchain technology for distributed storage, ensuring data security, integrity, and immutability, thereby enhancing user trust in the system. Through in-depth analysis of user search behavior, the system accurately understands user search preferences, providing a reliable foundation for personalized recommendations. By constructing intelligent user behavior profiles and personalized prediction models, the system can provide personalized search recommendations based on users' specific needs and interests, improving user satisfaction. Using deep neural network algorithms, the system can quickly and efficiently perform intelligent search recommendations on user input data, improving the relevance and accuracy of search results. Through continuous analysis of user search behavior and preferences, the system can continuously optimize the recommendation model, maintaining the efficiency and accuracy of the search engine.

[0054] Preferably, step S43 includes the following steps:

[0055] Step S431: Aggregate contextual information from user search behavior preference data to generate a user contextual information dataset;

[0056] Step S432: Perform time series analysis on the user context information dataset to generate user time series feature data;

[0057] Step S433: Perform user behavior analysis on the user context information dataset based on user time series feature data to generate user behavior data; construct user profiles based on user behavior data to generate user search behavior profiles;

[0058] Step S434: Perform multimodal feature fusion on the user context information dataset, user time series feature data, and user search behavior profile to generate a fused feature set; divide the fused feature set into a dataset to generate a model training set and a model test set;

[0059] Step S435: Use the DNN neural network algorithm to train the model on the training set to generate a personalized intelligent search recommendation training model; use the model test set to test and iterate the personalized intelligent search recommendation training model to generate a personalized intelligent search recommendation model.

[0060] This invention, through contextual information aggregation and time-series analysis, enables the system to construct more accurate and comprehensive user search behavior profiles, thereby improving recommendation accuracy. By fusing multimodal features, features from multiple data sources are combined, enhancing the model's understanding of user behavior and improving the performance of the recommendation system. Training the fused feature set using deep neural network algorithms allows the model to learn from complex user behavior patterns, improving recommendation effectiveness. Through iterative testing, the model can be continuously optimized and adjusted to improve recommendation accuracy. Based on user profiles and behavior analysis, the system can provide highly personalized search recommendations, improving user experience and the relevance of search results. Accurate and personalized recommendations not only improve the relevance of search results but also enhance user dependence on the system and satisfaction.

[0061] This specification provides an intelligent filtering system based on target search data, used to execute the aforementioned intelligent filtering method based on target search data. The intelligent filtering system based on target search data includes:

[0062] The search database integration module is used to acquire the original search dataset; preprocess the original search dataset to generate a standard search dataset; and construct a database based on the standard search dataset to obtain the search database.

[0063] The shallow search association module is used to acquire user input data; perform keyword parsing on the user input data to generate user input keyword data; perform search association on the search database based on the user input keyword data to generate a first search association dataset; compare the weight of the first search association data with a preset search association weight threshold; when the weight of the first search association data is greater than or equal to the preset search association weight threshold, sort and display the first search association dataset corresponding to the weight of the first search association data to generate first search association sorted data.

[0064] The deep search association module is used to assign attention weights to the first search association sorted data to generate attention weights for the first search association data; compare the attention weights of the first search association data with the preset standard data attention weight thresholds to obtain high search volume filtering data; calculate the similarity of the high search volume filtering data to obtain high search volume similarity; and perform a second filtering on the high search volume filtering data based on the high search volume similarity to obtain the second search association data.

[0065] The intelligent recommendation module is used to analyze user search behavior preferences using blockchain on the second search-related data, generating user search behavior preference data; constructing intelligent user behavior profiles based on user search behavior preference data, generating user search behavior profiles; and using deep neural network algorithms to perform intelligent search recommendations based on user search behavior profiles, generating a recommendation result set to perform intelligent filtering of target search data.

[0066] The beneficial effects of this invention lie in generating a standard search dataset by acquiring and preprocessing the original search dataset. This step helps extract and clean the original data to meet the needs of subsequent processing, ensuring data accuracy and consistency. After keyword parsing, user input data is used for related searches in the search database, generating a first search-related dataset. Based on a preset search-related weight threshold, the first search-related data is sorted and filtered to generate first search-related ranked data. This step helps match and rank user input with content in the search database, providing search results relevant to the user's query. By assigning attention weights to the first search-related ranked data, high-search-volume filtered data can be obtained. Then, similarity calculation is performed on the high-search-volume filtered data to further filter out second search-related data with high similarity. This step helps provide more relevant and accurate search results. By analyzing user search behavior preferences in the second search-related data, user search behavior preference data can be generated. Then, a deep neural network algorithm can be used to construct a user search behavior profile, and intelligent search recommendations can be made based on this profile, generating a recommendation result set. This step helps to provide personalized search recommendations that match users' preferences and needs, improving the search engine's performance and user experience. By processing and analyzing search data, and understanding and modeling user behavior, more accurate and personalized search results can be provided. Therefore, this invention improves the accuracy and intelligence of search results through data preprocessing, keyword parsing, attention weight allocation, and intelligent recommendation. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating the steps of an intelligent filtering method based on target search data.

[0068] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2.

[0069] Figure 3 for Figure 1 A detailed flowchart illustrating the implementation steps of step S3.

[0070] Figure 4 for Figure 1A detailed flowchart illustrating the implementation steps of step S4.

[0071] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0072] The technical method of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0073] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0074] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0075] To achieve the above objectives, please refer to Figures 1 to 4 A smart filtering method based on target search data, the method comprising the following steps:

[0076] Step S1: Obtain the original search dataset; preprocess the original search dataset to generate a standard search dataset; construct a database based on the standard search dataset to obtain the search database;

[0077] Step S2: Obtain user input data; perform keyword parsing on the user input data to generate user input keyword data; perform search association on the search database based on the user input keyword data to generate a first search association dataset; compare the weight of the first search association data with a preset search association weight threshold; when the weight of the first search association data is greater than or equal to the preset search association weight threshold, sort and display the first search association dataset corresponding to the weight of the first search association data to generate first search association sorted data;

[0078] Step S3: Assign attention weights to the first search-related sorted data to generate attention weights for the first search-related data; compare the attention weights of the first search-related data with the preset standard data attention weight thresholds to obtain high search volume filtering data; calculate the similarity of the high search volume filtering data to obtain high search volume similarity; perform secondary filtering on the high search volume filtering data based on the high search volume similarity to obtain second search-related data;

[0079] Step S4: Utilize blockchain to analyze user search behavior preferences in the second search-related data, generating user search behavior preference data; construct intelligent user behavior profiles based on user search behavior preference data, generating user search behavior profiles; utilize deep neural network algorithms to perform intelligent search recommendations based on user search behavior profiles, generating a recommendation result set to perform intelligent filtering of target search data.

[0080] This invention improves data quality and consistency by preprocessing and standardizing raw search data, thereby enabling more efficient construction of the search database. This contributes to improved search engine performance and accuracy. Keyword parsing and search association allow for the provision of relevant search results based on user input. Sorted display helps users find content of interest more quickly, enhancing the search experience. Attention weight allocation and similarity calculation further filter and refine search results, providing users with more relevant and accurate search recommendations. This increases user satisfaction and reduces search result redundancy. By analyzing user search behavior preferences and building behavioral profiles, the search engine can better understand user interests and needs. Combining deep neural network algorithms for intelligent search recommendations provides personalized search results, increasing user satisfaction and search efficiency. Simultaneously, blockchain technology ensures data security and trustworthiness, further enhancing the reliability of the search engine. Therefore, this invention improves the accuracy and intelligence of search results through data preprocessing, keyword parsing, attention weight allocation, and intelligent recommendation.

[0081] In this embodiment of the invention, reference Figure 1The above is a flowchart illustrating the steps of an intelligent filtering method based on target search data according to the present invention. In this example, the intelligent filtering method based on target search data includes the following steps:

[0082] Step S1: Obtain the original search dataset; preprocess the original search dataset to generate a standard search dataset; construct a database based on the standard search dataset to obtain the search database;

[0083] In this embodiment of the invention, search data is collected from multiple sources, such as web crawlers and API interfaces. The search dataset may include information such as user search keywords, search time, and clicked links. The data preprocessing stage includes steps such as data cleaning, deduplication, and normalization. For example, duplicate search records are removed, date and time formats are standardized to a uniform format, and missing values ​​are handled. This ensures data quality and consistency and prepares the data for subsequent analysis and modeling. A suitable database management system (such as MySQL, MongoDB, etc.) is used to design the database schema according to the structure of the standard search dataset. The preprocessed data is then imported into the database. When designing the database, the query requirements of the search data need to be considered so that subsequent search operations can be performed efficiently.

[0084] Step S2: Obtain user input data; perform keyword parsing on the user input data to generate user input keyword data; perform search association on the search database based on the user input keyword data to generate a first search association dataset; compare the weight of the first search association data with a preset search association weight threshold; when the weight of the first search association data is greater than or equal to the preset search association weight threshold, sort and display the first search association dataset corresponding to the weight of the first search association data to generate first search association sorted data;

[0085] In this embodiment of the invention, data is obtained through search engine search boxes, voice input, API interfaces, etc. This data includes user-input search keywords, query strings, voice commands, etc. Natural Language Processing (NLP) technology or keyword extraction algorithms are used to extract keywords from the user-input data. These keywords are key descriptions of the user's search intent or needs. The search engine's retrieval algorithm is used to retrieve relevant search results from the search database based on the user-input keywords. These search results include web pages, articles, images, etc., related to the user-input keywords. The search results are weighted according to metrics such as relevance, click-through rate, and pageviews. Then, they are compared with a preset search association weight threshold to filter out search results that meet the criteria, and then sorted and displayed according to their weights, generating the first search association ranking data.

[0086] Step S3: Assign attention weights to the first search-related sorted data to generate attention weights for the first search-related data; compare the attention weights of the first search-related data with the preset standard data attention weight thresholds to obtain high search volume filtering data; calculate the similarity of the high search volume filtering data to obtain high search volume similarity; perform secondary filtering on the high search volume filtering data based on the high search volume similarity to obtain second search-related data;

[0087] In this embodiment of the invention, an attention weight is assigned to each search result based on metrics such as relevance, authority, and click-through rate. These weights can be adjusted according to business needs and user behavior data to ensure that more relevant search results receive higher attention weights. A preset standard data attention weight threshold is set, and the attention weights of the first search-related data are compared. If the attention weight of a search result exceeds the preset threshold, it is identified as high-search-volume data. For the filtered high-search-volume data, their relevance can be further evaluated by calculating their similarity. Similarity calculation can employ text similarity algorithms, image similarity algorithms, etc., calculated based on the content characteristics of the search results. Based on the calculated similarity values, the high-search-volume data is further filtered to refine and optimize the search results. The data can be sorted according to similarity values, or a similarity threshold can be set to filter out the most relevant search results, generating second-related search data.

[0088] Step S4: Utilize blockchain to analyze user search behavior preferences in the second search-related data, generating user search behavior preference data; construct intelligent user behavior profiles based on user search behavior preference data, generating user search behavior profiles; utilize deep neural network algorithms to perform intelligent search recommendations based on user search behavior profiles, generating a recommendation result set to perform intelligent filtering of target search data.

[0089] In this embodiment of the invention, blockchain technology is used to ensure the security, immutability, and transparency of search behavior data. Search behavior data can include user clicks, favorites, shares, and other behaviors. Blockchain is used to record and store this data, ensuring its credibility. Then, this search behavior data is analyzed to extract user search preferences, such as preferred topics and keywords. Based on user search behavior preference data, a user behavior profile is constructed. This can include the user's areas of interest, search habits, and preference tags. Machine learning algorithms and natural language processing techniques can be used to analyze and model the search behavior data to generate the user behavior profile. Deep neural networks (DNNs) and other algorithms are used to model and analyze the user behavior profile to understand the user's search preferences and behavioral patterns. Then, based on the user's search behavior profile and the current search environment, recommendation algorithms are used to generate personalized search recommendations for the user. These recommendations can be articles, products, services, etc., related to the user's interests. The generated recommendation result set is used to intelligently filter target search data. Based on the user's search preferences and behavior profile, search results with high relevance to the user's interests are selected for display. This improves the relevance of search results and user satisfaction.

[0090] Preferably, step S1 includes the following steps:

[0091] Step S11: Obtain the original search dataset;

[0092] Step S12: Clean the original search dataset to generate the original search clean dataset; filter the original search clean dataset to generate the original search filtered dataset; identify the data format of the original search filtered dataset and standardize the data format of the identified original search filtered dataset to generate the search format dataset.

[0093] Step S13: Classify the search format dataset for duplicate data to obtain non-duplicate search data and duplicate search data, and remove duplicate search data;

[0094] Step S14: Perform missing value detection on the search non-duplicate data to generate missing value detection results for the search non-duplicate data; use the missing value detection results to repair the search non-duplicate data and generate a standard search dataset.

[0095] Step S15: Construct a database based on the standard search dataset to obtain the search database.

[0096] In this embodiment of the invention, search data is collected from various sources. The data can come from website logs, user behavior records, API call records, etc. The collected data includes search keywords, search time, search results, and other information. The data is cleaned to remove or repair erroneous, inconsistent, or invalid data. For example, this involves removing duplicate records, correcting format errors, and handling missing values. Data filtering is used to select data that meets specific conditions or rules, including deleting irrelevant data, removing noisy data, and filtering outliers. The data is converted to a unified format for subsequent processing and analysis, including converting data from different data types (such as text, date, number, etc.) to a unified format and standardizing the naming conventions of data fields. The cleaned and filtered data undergoes duplicate detection and classification. Duplicate data is identified and classified as duplicate datasets, while non-duplicate data is retained. Missing value detection is performed on the non-duplicate datasets to identify the location and type of missing values. Based on the missing value detection results, appropriate methods are used to repair or impute missing values. Common methods include using the mean, median, interpolation, etc., to impute missing values. Based on a processed standard search dataset, a search database is constructed using appropriate database technologies, involving operations such as creating database table structures, defining indexes, and setting data relationships to support efficient data storage, retrieval, and querying.

[0097] Preferably, step S14 includes the following steps:

[0098] Step S141: Use a conditional random field to detect missing values ​​in the search non-repeating data and obtain the missing data detection results for the search non-repeating data;

[0099] Step S142: Based on the search results for missing non-duplicate data, impute missing values ​​in the preliminary detected abnormal dataset to generate a search-impacted dataset; perform data consistency checks on the search-impacted data to obtain data consistency check results; and use the data consistency check results to identify data anomalies in the search-impacted dataset to generate data anomaly type identification results.

[0100] Step S143: When the data anomaly type identification result is confirmed to be a semantic anomaly, the knowledge graph method is used to perform semantic repair on the search-filled dataset to generate a semantic repair dataset.

[0101] Step S144: When the data anomaly type identification result is confirmed to be a statistical anomaly, Bayesian inference is used to perform statistical repair on the search-filled dataset to generate a statistically repaired dataset.

[0102] Step S145: Perform repair and verification on the semantic repair dataset and the statistical repair dataset to generate a standard search dataset.

[0103] In this embodiment of the invention, a Conditional Random Field (CRF) model is used to detect missing values ​​in the search non-repeating data. CRF is a probabilistic graphical model commonly used in sequence labeling tasks. Here, the search data can be considered as a sequence, and the CRF model is used to detect missing values ​​in the sequence, obtaining the missing value detection results for the search non-repeating data. Based on the missing value detection results, the initially detected abnormal dataset is imputed to generate a search imputed dataset. Consistency checks are performed on the imputed data to ensure internal consistency, such as ensuring the consistency of relationships between different fields and the consistency of data format. If the data consistency check confirms semantic anomalies, i.e., the data does not conform to the expected semantic rules, a knowledge graph method can be used to semantically repair the search imputed dataset. Knowledge graphs can provide semantic relationships and connection information between entities, used to repair semantic problems in the data and generate a semantically repaired dataset. If the data consistency check confirms statistical anomalies, i.e., the data does not statistically conform to the expected distribution pattern, a Bayesian inference method can be used to statistically repair the search imputed dataset. Bayesian inference can infer the values ​​of missing values ​​by modeling the probability distribution of data, thereby imputing them and generating a statistically repaired dataset. The dataset, after semantic and statistical repair, is then validated to ensure it meets the expected quality standards, generating the final standard search dataset.

[0104] As an example of the present invention, reference is made to... Figure 2 As shown, in this example, step S2 includes:

[0105] Step S21: Obtain user input data;

[0106] Step S22: Identify the input type of the user input data to obtain user input type data; perform keyword parsing on the user input data based on the user input type data to generate user input keyword data;

[0107] Step S23: Perform search association on the search database based on the user-input keyword data to generate a first search association dataset; calculate the association weights on the first search association dataset to obtain the weights of the first search association data;

[0108] Step S24: Compare the weight of the first search-related data with the preset search-related weight threshold. When the weight of the first search-related data is greater than or equal to the preset search-related weight threshold, sort and display the first search-related dataset corresponding to the weight of the first search-related data to generate the first search-related sorted data.

[0109] In this embodiment of the invention, input data is collected from the user, specifically in the form of text, voice, images, etc. Based on the data's features or metadata, the system uses machine learning algorithms or rule engines to identify the data type. For example, a text classification model can be used to identify text data, a speech recognition model to identify speech data, and an image classification model to identify image data. For text data, the system can use a word segmenter to break the text into words or phrases, and perform part-of-speech tagging and named entity recognition to identify keywords. For voice data, the system can use speech recognition technology to convert speech into text and then extract keywords. For image data, the system can use image processing and feature extraction techniques to identify key information in the image, such as identifying objects in the image using object detection algorithms. Search results are then searched and associated in a search database based on the user-input keyword data. Search results can be sorted based on factors such as keyword matching degree and relevance, and the association weight of each search result is calculated. The association weights in the search results are compared with a preset search association weight threshold. If the association weight of a search result is greater than or equal to the preset search association weight threshold, the search result is added to the final search result set, and the search results are sorted and displayed according to their association weights. The search results can be sorted in descending order based on relevance weights to ensure users see the most relevant results first. For example, if a user enters the text "car maintenance" into the search box, the text is identified as text and then segmented into keywords like "car" and "maintenance." The search database is then searched for relevant data containing these keywords. Relevance weights are calculated based on the relevance of the search results. Assume the search results are as follows:

[0110] Result 1: Title: "Car Maintenance Guide", Relevance: 0.8; Result 2: Title: "Frequently Asked Questions about Car Maintenance", Relevance: 0.7; Result 3: Title: "Car Repair Tips", Relevance: 0.5. For each result, a weight is assigned based on its relevance. For example, Result 1 has a weight of 0.8, Result 2 has a weight of 0.7, and Result 3 has a weight of 0.5. If the preset search relevance weight threshold is 0.6, then the weights of Result 1 and Result 2 meet the condition and can be added to the final search results set.

[0111] Preferably, step S22 includes the following steps:

[0112] Step S221: Perform input type recognition on the user input data to obtain user input type data, wherein the input type recognition includes text recognition, speech recognition, image recognition and video recognition;

[0113] Step S222: If the user input data is confirmed to be text, perform text semantic analysis on the user input data to generate text semantic analysis data; extract keywords from the user input data based on the text semantic analysis data to generate text keyword data.

[0114] Step S223: If the user input data is confirmed to be audio, perform a Fourier transform on the user input data to obtain the user input spectrogram; extract the Mel frequency cepstral coefficients from the user input spectrogram to obtain the user input audio features; transcribe the user input audio features into text to generate audio conversion keyword data.

[0115] Step S224: When it is confirmed that the user input type data is an image, image content recognition is performed on the user input data to generate image content recognition data, wherein image content recognition includes image object recognition and image scene recognition; the image content recognition data is used to describe the image content and generate image conversion keyword data.

[0116] Step S225: When it is confirmed that the user input data is video, the video frames are extracted from the user input data based on a preset time interval to obtain the user input video frame set; the frame set change is filtered in the user input video frame set to obtain the user input change frame set; the frame set image content is recognized in the user input change frame set to generate frame set image content data; the frame set image content data is described in the frame set image content data to generate video conversion keyword data;

[0117] Step S226: Perform comprehensive keyword extraction on the text keyword data, audio-to-keyword data, image-to-keyword data, and video-to-keyword data to obtain the user input keyword data.

[0118] In this embodiment of the invention, the input type is identified based on the features and metadata of the user input data. Specific input types include text, speech, image, and video. If the user input type is confirmed as text, the system performs the following processing: Text semantic analysis: Semantic analysis is performed on the text to understand its meaning and implications. Keyword extraction: Based on the text semantic analysis results, keywords are extracted as keyword data for the user input. If the user input type is confirmed as audio, the system performs the following processing: Fourier transform: Fourier transform is performed on the audio data to convert the audio signal into a spectrogram. Mel-frequency cepstral coefficient extraction: Mel-frequency cepstral coefficients are extracted from the spectrogram as audio features. Text transcription: The audio is transcribed into text to extract keywords as keyword data for the user input. If the user input type is confirmed as an image, the system performs the following processing: Image content recognition: Image content is recognized using image object recognition and image scene recognition technologies. Image content description: The image content is described based on the recognition results, and keywords are extracted as keyword data for the user input. If the user input data is confirmed to be video, the system performs the following processing: Video frame extraction: Frames are extracted from the video data based on preset time intervals to obtain a video frame set. Frame set change filtering: Frames in the video frame set that have changed are filtered out as the user input changed frame set. Frame set image content recognition: Image content recognition is performed on each frame in the changed frame set, and keywords are extracted as user input keyword data. Frame set image content description: Each frame in the frame set is described based on the recognition results, and keywords are extracted as user input keyword data. Finally, the system performs comprehensive keyword extraction from text keyword data, audio-to-video keyword data, image-to-video keyword data, and video-to-video keyword data to obtain the user input keyword data.

[0119] As an example of the present invention, reference is made to... Figure 3 As shown, step S3 in this example includes:

[0120] Step S31: Collect user feedback data from the first search related sorting data to obtain the first search user feedback data, wherein the feedback data collection includes user click frequency collection and user dwell time collection;

[0121] Step S32: Based on the feedback data from the first search user, assign attention weights to the first search-related ranking data to generate attention weights for the first search-related data;

[0122] Step S33: Compare the attention weight of the first search related data with the preset standard data attention weight threshold. When the attention weight of the first search related data is greater than or equal to the preset standard data attention weight threshold, the first search user feedback data is filtered for high search volume to obtain high search volume filtered data.

[0123] Step S34: Calculate the similarity of the high search volume filtering data using the search similarity calculation formula to obtain the high search volume similarity; perform a second filtering on the high search volume filtering data based on the high search volume similarity to obtain the second search related data.

[0124] In this embodiment of the invention, user feedback on the first search-related ranking data is collected, including user click frequency and user dwell time. For example, user click frequency data: {Result 1: 10 times, Result 2: 5 times, Result 3: 3 times}, and user dwell time data: {Result 1: 30 seconds, Result 2: 20 seconds, Result 3: 15 seconds}. Based on the first search user feedback data, the system assigns an attention weight to each search result to reflect the user's level of attention to the search result. The attention weight is calculated by weighting user click frequency and dwell time. Based on the attention weight of the first search-related data, the system filters out search results with higher attention for further processing. For example, the standard data attention weight threshold is 0.5, and the high search volume filtering data is: {Result 1: attention weight 0.7, Result 2: attention weight 0.6}. The similarity between search results is calculated based on the high search volume filtering data, and a second filtering is performed. Search similarity calculation formulas, such as cosine similarity or other similarity measurement methods, are used.

[0125] Preferably, the search similarity calculation formula in step S34 is as follows:

[0126]

[0127] In the formula, S(Q,D) represents the similarity between query Q and document D, a represents the lower limit of search time, b represents the upper limit of search time, n represents the number of queries, and w i Let λ represent the weight of the i-th selection feature. i Let q represent the attenuation coefficient of the i-th screening feature. i Let d represent the value of the i-th feature of Q. i Let exp(-λ) represent the value of the i-th feature of document D. i (q i -d i ) 2 ) represents the similarity between the i-th features, g(t) represents the contribution adjustment coefficient as it changes with time point t, and σ represents the standard deviation of the normal distribution.

[0128] This invention analyzes and integrates a search similarity calculation formula. The integral term in the formula accumulates similarity values ​​within the interval [a, b], capturing the similarity changes of queries and documents across different feature spaces and time dimensions. Through integration, dynamic or time-varying similarity calculations can be handled, adapting to different application scenarios. (Summation term) Summing over all n features comprehensively considers the contribution of each feature to the similarity score. This ensures that the similarity calculation covers all relevant features, providing a comprehensive similarity assessment. Weight w i This represents the weight of each feature i, adjusting its importance in similarity calculation. By adjusting the weights, important features can be prioritized, while the influence of less important features can be ignored or reduced, improving the accuracy and relevance of the calculation. The Gaussian function exp(-λ) i (q i -d i ) 2 )) represents the similarity between the i-th features, as q i and d i The difference increases and decreases exponentially. Parameter λ i This controls the degree of influence of feature differences and provides adjustment for the sensitivity of different features. The Gaussian function ensures that similarity is high when feature differences are small and decreases rapidly when differences are large, which is consistent with reality. The weight function g(t) can capture the change of feature similarity over time, adapting to the processing of dynamic data and time series data. Normalization constant. The normalization of similarity calculation results ensures they conform to a normal distribution, facilitating interpretation and comparison. Standardization reduces potential biases during calculation, providing stable similarity results. While conventional search similarity calculation formulas in the art yield the similarity between query Q and document D, the formula provided in this invention allows for a more accurate calculation of their similarity. The formula considers multiple features and their weights, integrating them through a Gaussian function and integrals to provide a comprehensive similarity assessment. The use of normalization constants and Gaussian functions ensures the stability and robustness of the similarity calculation, reducing the impact of noise and outliers. Through the synergistic effect of multiple parameters, accurate, comprehensive, and stable similarity assessments can be provided in high-volume search data processing, adapting to different application scenarios and needs.

[0129] As an example of the present invention, reference is made to... Figure 4 As shown, step S4 in this example includes:

[0130] Step S41: Use blockchain to perform distributed and trusted storage of the second search-related data, and generate second search-related storage data;

[0131] Step S42: Analyze user search behavior preferences based on the second search association storage data to generate user search behavior preference data;

[0132] Step S43: Construct an intelligent user behavior profile based on user search behavior preference data to generate a user search behavior profile; use a deep neural network algorithm to construct a personalized prediction model based on the user search behavior profile to generate a personalized intelligent search recommendation model.

[0133] Step S44: Based on the personalized intelligent search recommendation model, perform intelligent search recommendations on the user input data, generate a recommendation result set, and perform intelligent filtering of the target search data.

[0134] In this embodiment of the invention, a search engine is used to store second-related search data (including search records, click records, etc.) via blockchain technology. This means that the data is distributed across multiple nodes, each with a complete copy, thus ensuring data security and trustworthiness. The immutability and decentralization of blockchain prevent data tampering and single points of failure. The search engine analyzes the second-related search data stored on the blockchain, including user search keywords, clicked links, search times, and other information. Then, through data analysis and mining techniques, it performs preference analysis on user search behavior, such as determining user preferences for certain topics and frequently used search terms, and generates user search behavior preference data. The search engine uses this data to construct a user profile, including user interests, preferences, and behavioral habits. Then, using machine learning algorithms such as deep neural networks, it builds a personalized prediction model. This model can predict content that the user is interested in based on the user's profile and generate a personalized intelligent search recommendation model. Based on the personalized intelligent search recommendation model, it intelligently recommends search results based on the user's input. By analyzing the user's profile and preferences, it recommends search results related to the user's interests, thereby providing a more intelligent and personalized search experience.

[0135] Preferably, step S43 includes the following steps:

[0136] Step S431: Aggregate contextual information from user search behavior preference data to generate a user contextual information dataset;

[0137] Step S432: Perform time series analysis on the user context information dataset to generate user time series feature data;

[0138] Step S433: Perform user behavior analysis on the user context information dataset based on user time series feature data to generate user behavior data; construct user profiles based on user behavior data to generate user search behavior profiles;

[0139] Step S434: Perform multimodal feature fusion on the user context information dataset, user time series feature data, and user search behavior profile to generate a fused feature set; divide the fused feature set into a dataset to generate a model training set and a model test set;

[0140] Step S435: Use the DNN neural network algorithm to train the model on the training set to generate a personalized intelligent search recommendation training model; use the model test set to test and iterate the personalized intelligent search recommendation training model to generate a personalized intelligent search recommendation model.

[0141] In this embodiment of the invention, user search behavior preference data is aggregated with contextual information, including the time, location, and device information of the user's search, to generate a user contextual information dataset. This contextual information helps to better understand the background and intent behind the user's search behavior. Time-series analysis is performed on the user contextual information dataset, such as the trend of changes in user search frequency over time, to generate user time-series feature data. This feature data reflects the temporal relevance and changing patterns of user search behavior. Further analysis of the user contextual information dataset is performed based on the user's time-series feature data to generate user behavior data. Then, a user search behavior profile is constructed based on this behavior data, including the user's interests, preferences, and behavioral habits. The user contextual information dataset, time-series feature data, and search behavior profile are fused using multimodal features to generate a fused feature set. Then, the fused feature set is divided into a model training set and a model testing set for training and testing a personalized intelligent search recommendation model. Algorithms such as deep neural networks (DNNs) are used to train the model on the model training set to generate a training model for personalized intelligent search recommendations. Then, the trained model is tested and iterated using a model test set to optimize its performance, ultimately generating a personalized intelligent search recommendation model to provide users with personalized search recommendations. More specifically, for example, if a user performs multiple searches within a week, with different search results on weekdays and weekends, a dataset containing background information on user search behavior can be generated by aggregating these search behaviors and adding contextual information (such as search time, search location, and search device). Time-series analysis of user search behavior within a week is then performed, such as analyzing whether there are significant differences in search frequency between weekdays and weekends, or whether there are trends of increased or decreased user search behavior during specific time periods. These analyses can generate time-series feature data of the user. If it is found that users tend to search for leisure and entertainment content more on weekends, this behavioral pattern can be incorporated into the user's search behavior profile. The user's search behavior profile is then correlated with contextual information and time-series features. The fused feature set is then divided into training and testing datasets for building the personalized intelligent search recommendation model. Algorithms such as deep neural networks (DNNs) are used to train the model on the training set to generate a training model for personalized intelligent search recommendations. Then, the trained model is tested and iterated on a test set to optimize its performance. The final model can be used to provide users with personalized search recommendations.

[0142] This specification provides an intelligent filtering system based on target search data, used to execute the aforementioned intelligent filtering method based on target search data. The intelligent filtering system based on target search data includes:

[0143] The search database integration module is used to acquire the original search dataset; preprocess the original search dataset to generate a standard search dataset; and construct a database based on the standard search dataset to obtain the search database.

[0144] The shallow search association module is used to acquire user input data; perform keyword parsing on the user input data to generate user input keyword data; perform search association on the search database based on the user input keyword data to generate a first search association dataset; compare the weight of the first search association data with a preset search association weight threshold; when the weight of the first search association data is greater than or equal to the preset search association weight threshold, sort and display the first search association dataset corresponding to the weight of the first search association data to generate first search association sorted data.

[0145] The deep search association module is used to assign attention weights to the first search association sorted data to generate attention weights for the first search association data; compare the attention weights of the first search association data with the preset standard data attention weight thresholds to obtain high search volume filtering data; calculate the similarity of the high search volume filtering data to obtain high search volume similarity; and perform a second filtering on the high search volume filtering data based on the high search volume similarity to obtain the second search association data.

[0146] The intelligent recommendation module is used to analyze user search behavior preferences using blockchain on the second search-related data, generating user search behavior preference data; constructing intelligent user behavior profiles based on user search behavior preference data, generating user search behavior profiles; and using deep neural network algorithms to perform intelligent search recommendations based on user search behavior profiles, generating a recommendation result set to perform intelligent filtering of target search data.

[0147] The beneficial effects of this invention lie in generating a standard search dataset by acquiring and preprocessing the original search dataset. This step helps extract and clean the original data to meet the needs of subsequent processing, ensuring data accuracy and consistency. After keyword parsing, user input data is used for related searches in the search database, generating a first search-related dataset. Based on a preset search-related weight threshold, the first search-related data is sorted and filtered to generate first search-related ranked data. This step helps match and rank user input with content in the search database, providing search results relevant to the user's query. By assigning attention weights to the first search-related ranked data, high-search-volume filtered data can be obtained. Then, similarity calculation is performed on the high-search-volume filtered data to further filter out second search-related data with high similarity. This step helps provide more relevant and accurate search results. By analyzing user search behavior preferences in the second search-related data, user search behavior preference data can be generated. Then, a deep neural network algorithm can be used to construct a user search behavior profile, and intelligent search recommendations can be made based on this profile, generating a recommendation result set. This step helps to provide personalized search recommendations that match users' preferences and needs, improving the search engine's performance and user experience. By processing and analyzing search data, and understanding and modeling user behavior, more accurate and personalized search results can be provided. Therefore, this invention improves the accuracy and intelligence of search results through data preprocessing, keyword parsing, attention weight allocation, and intelligent recommendation.

[0148] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0149] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. An intelligent filtering method based on target search data, characterized in that, Includes the following steps: Step S1: Obtain the original search dataset; preprocess the original search dataset to generate a standard search dataset; construct a database based on the standard search dataset to obtain the search database; Step S2: Obtain user input data; perform keyword parsing on the user input data to generate user input keyword data; perform search association on the search database based on the user input keyword data to generate a first search association dataset; compare the weight of the first search association data with a preset search association weight threshold; when the weight of the first search association data is greater than or equal to the preset search association weight threshold, sort and display the first search association dataset corresponding to the weight of the first search association data to generate first search association sorted data; Step S3: Assign attention weights to the first search-related sorted data to generate attention weights for the first search-related data; The attention weights of the first search-related data are compared with the preset standard data attention weight thresholds to obtain high-search-volume filtering data; similarity is calculated on the high-search-volume filtering data to obtain high-search-volume similarity; the high-search-volume filtering data is then filtered a second time based on the high-search-volume similarity to obtain the second search-related data. Step S4: Utilize blockchain to analyze user search behavior preferences in the second search-related data, and generate user search behavior preference data; Intelligent user behavior profiling is constructed based on user search behavior preference data to generate user search behavior profiles; By using deep neural network algorithms to create user search behavior profiles, intelligent search recommendations are made and a set of recommended results is generated to perform intelligent filtering of target search data. Step S3 includes the following steps: Step S31: Collect user feedback data from the first search related sorting data to obtain the first search user feedback data, wherein the feedback data collection includes user click frequency collection and user dwell time collection; Step S32: Based on the feedback data from the first search user, assign attention weights to the first search-related ranking data to generate attention weights for the first search-related data; Step S33: Compare the attention weight of the first search related data with the preset standard data attention weight threshold. When the attention weight of the first search related data is greater than or equal to the preset standard data attention weight threshold, the first search user feedback data is filtered for high search volume to obtain high search volume filtered data. Step S34: Calculate the similarity of the high search volume filtering data using the search similarity calculation formula to obtain the high search volume similarity; perform a second filtering on the high search volume filtering data based on the high search volume similarity to obtain the second search related data; The search similarity calculation formula in step S34 is as follows: In the formula, S(Q,D) represents the similarity between query Q and document D, a represents the lower limit of search time, b represents the upper limit of search time, n represents the number of queries, and w i Let λ represent the weight of the i-th selection feature. i Let q represent the attenuation coefficient of the i-th screening feature. i Let d represent the value of the i-th feature of Q. i Let exp(-λ) represent the value of the i-th feature of document D. i (q i -d i ) 2 ) represents the similarity between the i-th features, g(t) represents the contribution adjustment coefficient as it changes with time point t, and σ represents the standard deviation of the normal distribution.

2. The intelligent filtering method based on target search data according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain the original search dataset; Step S12: Clean the original search dataset to generate the original search clean dataset; filter the original search clean dataset to generate the original search filtered dataset; identify the data format of the original search filtered dataset and standardize the data format of the identified original search filtered dataset to generate the search format dataset. Step S13: Classify the search format dataset for duplicate data to obtain non-duplicate search data and duplicate search data, and remove duplicate search data; Step S14: Perform missing value detection on the search non-duplicate data to generate missing value detection results for the search non-duplicate data; use the missing value detection results to repair the search non-duplicate data and generate a standard search dataset; Step S15: Construct a database based on the standard search dataset to obtain the search database.

3. The intelligent filtering method based on target search data according to claim 2, characterized in that, Step S14 includes the following steps: Step S141: Use a conditional random field to detect missing values ​​in the search non-repeating data and obtain the missing data detection results for the search non-repeating data; Step S142: Based on the search results for missing non-duplicate data, impute missing values ​​in the preliminary detected abnormal dataset to generate a search-impacted dataset; perform data consistency checks on the search-impacted data to obtain data consistency check results; and use the data consistency check results to identify data anomalies in the search-impacted dataset to generate data anomaly type identification results. Step S143: When the data anomaly type identification result is confirmed to be a semantic anomaly, the knowledge graph method is used to perform semantic repair on the search-filled dataset to generate a semantic repair dataset. Step S144: When the data anomaly type identification result is confirmed to be a statistical anomaly, Bayesian inference is used to perform statistical repair on the search-filled dataset to generate a statistically repaired dataset. Step S145: Perform repair and verification on the semantic repair dataset and the statistical repair dataset to generate a standard search dataset.

4. The intelligent filtering method based on target search data according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Obtain user input data; Step S22: Identify the input type of the user input data to obtain user input type data; perform keyword parsing on the user input data based on the user input type data to generate user input keyword data; Step S23: Perform search association on the search database based on the user-input keyword data to generate a first search association dataset; calculate the association weights on the first search association dataset to obtain the weights of the first search association data; Step S24: Compare the weight of the first search-related data with the preset search-related weight threshold. When the weight of the first search-related data is greater than or equal to the preset search-related weight threshold, sort and display the first search-related dataset corresponding to the weight of the first search-related data to generate the first search-related sorted data.

5. The intelligent filtering method based on target search data according to claim 4, characterized in that, Step S22 includes the following steps: Step S221: Perform input type recognition on the user input data to obtain user input type data, wherein the input type recognition includes text recognition, speech recognition, image recognition and video recognition; Step S222: If the user input data is confirmed to be text, perform text semantic analysis on the user input data to generate text semantic analysis data; extract keywords from the user input data based on the text semantic analysis data to generate text keyword data. Step S223: If the user input data is confirmed to be audio, perform a Fourier transform on the user input data to obtain the user input spectrogram; extract the Mel frequency cepstral coefficients from the user input spectrogram to obtain the user input audio features; transcribe the user input audio features into text to generate audio conversion keyword data. Step S224: When it is confirmed that the user input type data is an image, image content recognition is performed on the user input data to generate image content recognition data, wherein image content recognition includes image object recognition and image scene recognition; the image content recognition data is used to describe the image content and generate image conversion keyword data. Step S225: When it is confirmed that the user input data is video, the video frames are extracted from the user input data based on a preset time interval to obtain the user input video frame set; the frame set change is filtered in the user input video frame set to obtain the user input change frame set; the frame set image content is recognized in the user input change frame set to generate frame set image content data; the frame set image content data is described in the frame set image content data to generate video conversion keyword data; Step S226: Perform comprehensive keyword extraction on the text keyword data, audio-to-keyword data, image-to-keyword data, and video-to-keyword data to obtain the user input keyword data.

6. The intelligent filtering method based on target search data according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Use blockchain to perform distributed and trusted storage of the second search-related data, and generate second search-related storage data; Step S42: Analyze user search behavior preferences based on the second search association storage data to generate user search behavior preference data; Step S43: Construct an intelligent user behavior profile based on user search behavior preference data to generate a user search behavior profile; use a deep neural network algorithm to construct a personalized prediction model based on the user search behavior profile to generate a personalized intelligent search recommendation model. Step S44: Based on the personalized intelligent search recommendation model, perform intelligent search recommendations on the user input data, generate a recommendation result set, and perform intelligent filtering of the target search data.

7. The intelligent filtering method based on target search data according to claim 6, characterized in that, Step S43 includes the following steps: Step S431: Aggregate contextual information from user search behavior preference data to generate a user contextual information dataset; Step S432: Perform time series analysis on the user context information dataset to generate user time series feature data; Step S433: Perform user behavior analysis on the user context information dataset based on user time series feature data to generate user behavior data; construct user profiles based on user behavior data to generate user search behavior profiles; Step S434: Perform multimodal feature fusion on the user context information dataset, user time series feature data, and user search behavior profile to generate a fused feature set; divide the fused feature set into a dataset to generate a model training set and a model test set; Step S435: Use the DNN neural network algorithm to train the model on the training set to generate a personalized intelligent search recommendation training model; use the model test set to test and iterate the personalized intelligent search recommendation training model to generate a personalized intelligent search recommendation model.

8. An intelligent filtering system based on target search data, characterized in that, For executing the intelligent filtering method based on target search data as described in claim 1, the intelligent filtering system based on target search data comprises: The search database integration module is used to acquire the original search dataset; preprocess the original search dataset to generate a standard search dataset; and construct a database based on the standard search dataset to obtain the search database. The shallow search association module is used to acquire user input data; perform keyword parsing on the user input data to generate user input keyword data; perform search association on the search database based on the user input keyword data to generate a first search association dataset; compare the weight of the first search association data with a preset search association weight threshold; when the weight of the first search association data is greater than or equal to the preset search association weight threshold, sort and display the first search association dataset corresponding to the weight of the first search association data to generate first search association sorted data. The deep search association module is used to assign attention weights to the first search association sorted data to generate attention weights for the first search association data; compare the attention weights of the first search association data with the preset standard data attention weight thresholds to obtain high search volume filtering data; calculate the similarity of the high search volume filtering data to obtain high search volume similarity; and perform a second filtering on the high search volume filtering data based on the high search volume similarity to obtain the second search association data. The intelligent recommendation module is used to analyze user search behavior preferences using blockchain on the second search-related data, generating user search behavior preference data; constructing intelligent user behavior profiles based on user search behavior preference data, generating user search behavior profiles; and using deep neural network algorithms to perform intelligent search recommendations based on user search behavior profiles, generating a recommendation result set to perform intelligent filtering of target search data.

Images