Personalized information search system for different types of barrier groups

Abstract

The invention relates to the technical field of information retrieval, in particular to a personalized information search system for different types of barrier groups, which comprises a voice text conversion module, a visual text synthesis module, a semantic fusion and retrieval module and a user targeted recommendation module. According to the method, the voice recognition precision is greatly improved by analyzing the audio signal and extracting the key phoneme sequence, especially in a noisy environment, the system can provide richer text information for visually impaired users through deep analysis of picture data, the accessibility of information is greatly improved by converting visual elements into text descriptions, and the user experience is improved. Through high matching and sorting of text information and continuous optimization of the search process, a search engine can respond to query of a user more accurately, information redundancy is reduced, search results are optimized by analyzing historical behaviors of the user, the user can obtain related information more quickly, and the user experience is improved. Therefore, the overall search efficiency and the user satisfaction degree are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of information retrieval, and in particular to a personalized information search system for different types of handicapped groups. Background Art

[0002] Information retrieval technology involves the design, implementation, and evaluation of systems that enable users to effectively retrieve and obtain information. This field includes not only the development of search engines and database query languages, but also covers the application of technologies such as data mining, natural language processing, and machine learning to improve the relevance and accuracy of retrieval. The main goal of an information retrieval system is to quickly and accurately find the information users need from large amounts of unstructured or semi-structured data.

[0003] Among them, the personalized information search system for different types of disabled groups is a specific information retrieval system that aims to provide customized search solutions for different types of disabled groups. This system understands and adapts to the specific needs and preferences of each user, and uses algorithms to adjust search results and interface design to improve information accessibility for disabled user groups. Its uses include helping the visually impaired, hearing-impaired or other people with special needs to more easily find helpful information on the Internet, thereby enhancing their autonomy and improving their quality of life.

[0004] The existing technology ignores the adaptive design for special user groups, especially when processing data input from visually impaired or hearing-impaired users, and fails to fully optimize the recognition range of voice commands and the ability to interpret image information, resulting in users from disabled groups facing problems such as inaccurate information acquisition or complex operations. In the existing technology, the semantic fusion of multiple data sources is not sufficient, and there is a lack of effective targeted search strategies, which makes users often receive search results that do not match their needs. In addition, the existing technology has obvious shortcomings in the ability to dynamically adjust the search process to adapt to the changing needs of users, which limits its application effect and user satisfaction in special situations. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a personalized information search system for different types of handicapped groups.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a personalized information search system for different types of handicapped groups, the system comprising: The speech-to-text conversion module analyzes the spectral characteristics of the audio signal based on the voice command issued by the user, calculates the energy peak and resonance peak position of the signal, identifies the energy distribution of multiple frequency bands, and extracts the key phoneme sequence, matches it with the grammatical rules, and then converts it into structured text to obtain the speech data analysis result; The visual text synthesis module collects the image data uploaded by users, identifies the objects and scenes in the image, extracts the text information in the image, analyzes the color, shape and size parameters, and converts the identified visual information into text description to obtain the visual information encoding result; The semantic fusion and retrieval module performs text information matching and sorting based on the speech data analysis results and visual information encoding results, analyzes the semantic relevance of different data sources, optimizes the information retrieval process, and corrects information deviations to obtain a semantic index configuration; The user-targeted recommendation module analyzes the user's search behavior and preference data based on the semantic index configuration, executes search queries, filters the search results of the queries, extracts information related to user needs, and automatically processes content summaries, and then sorts and optimizes the display of the summary information to obtain a search optimized summary.

[0007] The present invention is improved in that the step of identifying the energy distribution of the multi-bands is specifically as follows: Based on the voice commands issued by the user, noise reduction and standardization are performed, and the time domain signal is converted into a frequency domain signal using fast Fourier transform to obtain spectrum data; Based on the spectrum data, the formula is adopted: ; Measure the energy density within the frequency band to obtain the energy peak and resonance peak position of each frequency band, where: is at frequency The energy value below is the sample size, is at frequency and sample index The complex output of the FFT result is yes The module length; Based on the energy peak and resonance peak position of each frequency band, the audio spectrum is analyzed to divide it into low frequency, medium frequency and high frequency bands, and the total energy of each frequency band is calculated to obtain the energy distribution of multiple frequency bands.

[0008] The present invention is improved in that the steps of obtaining the speech data analysis result are specifically as follows: Based on the energy distribution of the multiple frequency bands and in combination with the spectrum data features, phoneme recognition is performed, and a preliminary phoneme recognition result is obtained by comparing the frequency of the energy peak in the frequency band with the spectrum features of the known phonemes; Based on the preliminary phoneme recognition results, the formula is adopted: ; Re-evaluate the matching degree between the phoneme sequence and the grammatical rules, verify the correctness of the word order, and adjust or replace the unmatched phonemes to obtain the grammatical matching result, where: is the posterior probability, indicating that when the phoneme sequence is detected Under the condition of The probability of occurrence, is the likelihood probability, expressed in phonemes Under the condition of occurrence, the phoneme sequence is detected The probability of It is a phoneme The prior probability of is a phoneme sequence The marginal probability of Based on the grammatical matching result, the phoneme sequence is converted into a structured text output, including translating the phonemes into corresponding texts, and applying linguistic rules to optimize the semantic level to obtain the speech data analysis result.

[0009] The present invention is improved in that the analysis steps of the color, shape and size parameters are specifically as follows: Collect the image data uploaded by users, process the uploaded images, including color balance, brightness adjustment and contrast optimization, and obtain optimized image data; Based on the optimized image data, the color, shape and size parameters of multiple objects in the image are calculated using the formula: ; Get the feature score of each object, where is the average color value of the object, is the shape complexity, is the size parameter, is the color weight coefficient, is the shape weight coefficient, is the size weight coefficient; Based on the feature score of each object, the objects are classified and sorted, and the objects with key scores are processed first to obtain the color, shape and size parameter analysis results.

[0010] The present invention is improved in that the step of obtaining the visual information encoding result is specifically as follows: Based on the color, shape and size parameters, matching is performed with the visual information template to determine the classification label of each object and obtain a preliminary labeling result; Based on the preliminary labeling result, it is converted into a text description, the fluency and accuracy of the text description are adjusted, and the text description is encoded into a data format to obtain a visual information encoding result.

[0011] The present invention is improved in that the analysis step of the semantic relevance of the difference data source is specifically as follows: Based on the speech data analysis results and the visual information encoding results, extract key semantic tags for each data source to obtain two independent key semantic tag lists; The two independent key semantic tag lists are cross-compared using the formula: ; Compute the normalized cosine similarity between label vectors , and get the semantic similarity score, where Representative The weight of key semantic tags, Representatives and The semantic score of the matching tags, Represents the total number of key semantic tags; Based on the semantic similarity score, the relevance between the different data sources is evaluated, and the key related contents are marked to obtain semantic relevance data.

[0012] The present invention is improved in that the step of obtaining the semantic index configuration is specifically as follows: Based on the semantic relevance, an index infrastructure is established, including keywords and their connection strengths, to obtain a preliminary retrieval framework; Based on the preliminary retrieval framework, the information retrieval process is optimized, and the information deviation in the index is dynamically adjusted and corrected through query testing and user feedback to obtain a semantic index configuration.

[0013] The present invention is improved in that the step of obtaining the search optimization summary is specifically as follows: Based on the semantic index configuration, keyword analysis is performed, and according to the user's historical behavior and preference data, a targeted search query is executed to obtain a preliminary search result set; Based on the preliminary search result set, the formula is used: ; Calculate the score for each search result , and automatically process the content summary to obtain an optimized content summary, where Indicates Visibility score for search results, Indicates that for User personalized weight of search results, is the number of search results; The optimized content summaries are sorted and displayed in an optimized manner, and the display method and order are adjusted according to the user's preference to obtain a search optimized summary.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by analyzing audio signals and extracting key phoneme sequences, the accuracy of speech recognition is greatly improved, especially in noisy environments. The in-depth analysis of image data enables the system to provide visually impaired users with richer text information. By converting visual elements into text descriptions, the accessibility of information is greatly improved. The high degree of matching and sorting of text information and the continuous optimization of the search process enable the search engine to respond to user queries more accurately, reduce information redundancy, and optimize search results by analyzing user historical behaviors, so that users can obtain relevant information more quickly, thereby improving overall search efficiency and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a system flow chart of the present invention; Figure 2 A flowchart for identifying the energy distribution of multiple frequency bands in the present invention; Figure 3 A flowchart of obtaining the speech data analysis result in the present invention; Figure 4 is a flow chart for analyzing color, shape and size parameters in the present invention; Figure 5 A flowchart of obtaining the visual information encoding result in the present invention; Figure 6 It is a flowchart of analyzing the semantic relevance of the difference data sources in the present invention; Figure 7 A flowchart for obtaining the semantic index configuration in the present invention; Figure 8 The present invention is a flowchart for obtaining search optimization abstracts. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating positions or positional relationships, are based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0018] Example See also Figure 1 The present invention provides a technical solution: a personalized information search system for different types of handicapped groups, comprising: The speech-to-text conversion module analyzes the spectral characteristics of the audio signal based on the voice command issued by the user, calculates the energy peak and resonance peak position of the signal, identifies the energy distribution of multiple frequency bands, and extracts the key phoneme sequence, matches it with the grammatical rules, and then converts it into structured text to obtain the speech data analysis result; The visual text synthesis module collects the image data uploaded by users, identifies the objects and scenes in the image, extracts the text information in the image, analyzes the color, shape and size parameters, and converts the identified visual information into text description to obtain the visual information encoding result; The semantic fusion and retrieval module performs text information matching and sorting based on the speech data analysis results and visual information encoding results, analyzes the semantic relevance of different data sources, optimizes the information retrieval process, and corrects information deviations to obtain semantic index configuration; The user-targeted recommendation module is based on semantic index configuration, analyzes user search behavior and preference data, executes search queries, filters the search results, extracts information related to user needs, and automatically processes content summaries, and then sorts and optimizes the display of summary information to obtain a search-optimized summary.

[0019] The results of speech data analysis include speech feature data, phoneme data, and grammatical matching output results. The results of visual information encoding include object recognition labels, scene description text, color and shape analysis results. The semantic index configuration includes semantic relevance scores, data source difference analysis results, and retrieval process correction indicators. The search optimization summary includes preference matching analysis results, key information screening results, and summary text sorting results.

[0020] See also Figure 2 , the specific steps for identifying the energy distribution of multiple frequency bands are: Based on the voice commands issued by the user, noise reduction and standardization are performed, and the time domain signal is converted into a frequency domain signal using fast Fourier transform to obtain spectrum data; First, the audio signal is collected and preprocessed. The frequency domain filtering technology is used to remove the low-frequency background noise and high-frequency interference components in the signal segment by segment. The signal segments are extracted after dividing the time window through short-time Fourier transform, and the signal amplitude is normalized to the standardized range. The normalization operation calculates the maximum signal value in each time window and scales all amplitudes by the inverse of the maximum value to ensure that the input data range of subsequent signal processing steps is consistent. Then, the fast Fourier transform is used to convert the time domain signal into a frequency domain signal, so as to obtain the frequency distribution and amplitude information in the signal spectrum and obtain the spectrum data.

[0021] Based on the spectrum data, the formula is used: ; Measure the energy density within the frequency band to obtain the energy peak and resonance peak position of each frequency band, where: is at frequency The energy value under is used to indicate the signal strength at a specific frequency point. is the sample size, is at frequency and sample index The complex output of the FFT result below represents the amplitude and phase information of the frequency and sample point. yes The modulus length, that is, at the frequency and sample index The signal amplitude under Indicates the last index of the sample sequence, index Starting from 0, to Finish; If the frequency , the number of sampling points is , the two FFT amplitude values ​​are , , substitute into the formula: ; Compute the square of the magnitude of each value: ; ; Sum: ; Calculate the average: ; The result shows that the energy density at a frequency of 1000 Hz is 3.125, which is further used to analyze the energy peak and resonance peak position within the frequency band and provide a basis for subsequent spectrum segmentation and energy distribution.

[0022] Based on the energy peak and resonance peak position of each frequency band, the audio spectrum is analyzed to divide it into low frequency, medium frequency and high frequency bands, and the total energy of each frequency band is calculated to obtain the energy distribution of multiple frequency bands; The spectrum is divided into low-frequency band, mid-frequency band and high-frequency band. The low-frequency band is distinguished by fundamental frequency and harmonic calculation. The harmonic frequency range is determined by the multiple relationship of the fundamental frequency. The mid-frequency band is the range from the fundamental frequency and harmonics to the preset high-frequency threshold, and the high-frequency band is the remaining spectrum area. Then the energy of the frequency points in each frequency band is calculated. During the calculation process, the total energy is obtained by summing the amplitude squares of the frequency components in the frequency band and taking the average value. Finally, the total energy is summarized and organized into multi-band energy distribution.

[0023] See also Figure 3 , the specific steps for obtaining the speech data analysis results are: Based on the energy distribution of multiple frequency bands and combined with the characteristics of spectrum data, phoneme recognition is performed. By comparing the frequency of energy peaks in the frequency band with the spectrum characteristics of known phonemes, preliminary phoneme recognition results are obtained. First, the spectral data features are extracted. The frequency band is divided into several energy distribution units by segmenting the frequency band, and the energy peak position of each unit and its corresponding frequency are extracted. Then, they are compared item by item with the known phoneme spectral feature database, which contains the typical frequency feature range and amplitude threshold of phonemes. During the comparison, the frequency and amplitude characteristics of the frequency band energy peak are matched and analyzed to calculate the matching degree and screen out the phoneme features that meet the conditions. By comprehensively combining the phoneme feature information with a high matching degree, a preliminary phoneme recognition result is generated.

[0024] Based on the preliminary phoneme recognition results, the formula is adopted: ; Re-evaluate the matching degree between the phoneme sequence and the grammatical rules, verify the correctness of the word order, and adjust or replace the unmatched phonemes to obtain the grammatical matching result, where: is the posterior probability, indicating that when the phoneme sequence is detected Under the condition of The probability of occurrence, is the likelihood probability, expressed in phonemes Under the condition of occurrence, the phoneme sequence is detected The probability of It is a phoneme The prior probability of the phoneme before monitoring any specific speech data The probability of occurrence, is a phoneme sequence The marginal probability of all phonemes can be regarded as Leading to monitoring The sum of the probabilities of Re-evaluate the matching degree between the phoneme sequence and the grammatical rules. During the evaluation process, the grammatical rule library is used to match the phoneme sequence item by item. First, the grammatical pattern that the current sequence conforms to is selected from the grammatical rule library, the audio features of each phoneme are re-marked, and the order of the phonemes is adjusted according to the requirements of the grammatical rules. Then, the comprehensive matching degree between the adjusted phoneme sequence and the grammatical rules is calculated. The phonemes that do not meet the grammatical matching conditions are marked as errors and replaced with the phonemes with the highest matching degree in the rule library. The above operations complete the grammatical correctness verification of the phoneme sequence and generate the grammatical matching results. If there is the following data, , by analyzing the phonemes The characteristics and observed The matching degree statistics show that , calculated directly by the frequency of occurrence of phonemes in the corpus, , by calculating all phoneme combinations leading to the observation The sum of the probabilities is obtained and substituted into the formula for calculation: ; This result shows that phonemes When observing the phoneme sequence The matching probability is 0.64, or 64%. This result is used to verify the correctness of the grammatical rules and to dynamically adjust the unmatched phonemes.

[0025] Based on the grammatical matching results, the phoneme sequence is converted into structured text output, including translating the phonemes into corresponding texts and applying linguistic rules to optimize the semantic level to obtain the speech data analysis results; The phoneme sequence is converted into structured text output. During the conversion process, the frequency features in the phoneme sequence are mapped to the corresponding text encoding area. The mapping relationship between phonemes and text is provided by the linguistic rule library. Through this rule library, a one-to-one or one-to-many correspondence is established between the frequency and amplitude features of the phonemes. Subsequently, the text structure is sorted and optimized, including adjusting the sentence order according to semantic rules, correcting semantic ambiguity, and segmenting the structured text. Finally, text data with a clear semantic hierarchy is output to obtain the speech data analysis result.

[0026] See also Figure 4 , the analysis steps of color, shape and size parameters are as follows: Collect the image data uploaded by users, process the uploaded images, including color balance, brightness adjustment and contrast optimization, and obtain optimized image data; Collect image data uploaded by users, perform color balance processing on the images, adjust the color value of the pixel by calculating the difference between the color distribution of each pixel and the standard color model, and use the histogram equalization method to calculate the pixel distribution of each brightness level during the brightness adjustment process, and correct the brightness range through normalization. During the contrast optimization process, the pixel value is remapped to the preset range through the contrast stretching method, thereby enhancing the contrast and detail expression of the image and generating optimized image data.

[0027] Based on the optimized image data, the color, shape and size parameters of multiple objects in the image are calculated using the formula: ; Get the feature score of each object, where is the average color value of the object, calculated as the average of the colors of all pixels in the image or image region, It is the shape complexity, which is a parameter to measure the complexity of the shape of an object. It can be calculated by the geometric features of the object, such as edge length, number of corner points, or the ratio of edge to area. is a size parameter, which is the physical size of the object, such as length, width, or area, is the color weight coefficient, is the shape weight coefficient, is the size weight coefficient; The multi-object area is isolated through the image segmentation method, and the features of the pixels of each object are extracted. First, the average color value of the object is calculated according to the RGB model, and the color value of each pixel is accumulated in turn and divided by the total number of pixels. Then, the edge contour of the object is obtained by the edge detection method, and the shape complexity of the object is calculated by the ratio of the edge length to the area of ​​the region. Finally, the maximum length and width of the object are measured to calculate its size parameters. Combined with the analysis results, a comprehensive feature score for each object is generated. If the following data is collected: , , , (Calculated from the color value of the object pixel, which is the RGB value); (calculated by edge length 150 and area 100); (Calculated from the object's length 10 and width 5), substitute into the formula: ; ; The result indicates that the object has a feature score of 63.45, which can be used for subsequent object classification and sorting to ensure that the analysis can prioritize key objects.

[0028] Based on the feature score of each object, objects are classified and sorted, and objects with key scores are prioritized to obtain color, shape and size parameter analysis results; Analyze the color features of the object and extract the average color value. The value is determined by counting the RGB values ​​of each object pixel and calculating its arithmetic mean. The shape complexity is calculated by detecting the outline of the object edge and combining the ratio of the edge length to the object area. The size parameter is determined by detecting the bounding box of the object and calculating the maximum length and width. The color, shape and size feature scores are then weighted and combined according to the preset weight coefficients. The objects are sorted according to the feature scores, and objects with higher scores are marked as key objects first. Finally, the color, shape and size parameter analysis results are generated.

[0029] See also Figure 5 , the steps for obtaining the visual information encoding results are as follows: Based on color, shape and size parameters, match with visual information templates to determine the classification label of each object and obtain preliminary labeling results; The parameters are matched one by one with the classification standards in the visual information template. The color distribution of each object is given priority in the matching process. The color matching degree is determined by calculating the Euclidean distance between the average value of the main color of each object and the standard color value in the template. At the same time, the shape parameters, such as the comparison value of edge complexity and the template shape standard, are combined to evaluate the shape similarity. Finally, the size parameter is used to compare the area of ​​the object with the size range defined in the template to complete the final classification match. After all parameters are matched, the object will be assigned the most suitable classification label to obtain a preliminary labeling result.

[0030] Based on the preliminary labeling results, convert them into text descriptions, adjust the fluency and accuracy of the text descriptions, and encode the text descriptions into data formats to obtain visual information encoding results; Extract the classification labels and corresponding parameter information from the labeling results, and convert the classification labels of each object into natural language text descriptions. The generation of text descriptions requires comprehensive classification characteristics of color, shape, and size. Then, perform grammatical analysis and fluency optimization on the generated descriptions, including adjusting the sentence structure, correcting inconsistencies in wording, and eliminating semantic repetition or ambiguity to ensure that the generated text descriptions are clear and accurate. Finally, the optimized text descriptions are encoded and stored in JSON format to obtain the visual information encoding results.

[0031] See also Figure 6 ,The specific steps for analyzing the semantic relevance of the ,different data sources are as follows: Based on the speech data analysis results and visual information encoding results, key semantic labels are extracted for each data source to obtain two independent key semantic label lists; By parsing the text content in the speech data and the descriptive content in the visual information encoding results, semantic analysis is performed on the two data sources respectively, and key semantic elements with higher weights are extracted. The weights are calculated based on the frequency of occurrence and association strength of the labels in their corresponding contexts. The extracted key semantic labels are then sorted by weight and classified into two independent semantic label lists according to their sources, resulting in two independent key semantic label lists.

[0032] Cross-comparison is performed between two independent lists of key semantic tags using the formula: ; Compute the normalized cosine similarity between label vectors , and get the semantic similarity score, where Representative The weight of a key semantic tag is the importance or frequency of the tag in the source data. Representatives and The semantic score of the matching tags is a score obtained based on some kind of semantic analysis or similarity comparison. Represents the total number of key semantic tags; The following data were collected: =0.4, =0.3, =0.3 represent the weights of the three key semantic tags, which are obtained by analyzing their frequency and contextual relevance in the source data; =0.9, =0.85, =0.8 are the semantic scores matching the three tags respectively; =3The total number of key semantic tags, determined by the number of extracted tags.

[0033] Molecular calculation: ; ; Weight vector norm: ; ; Semantic score vector modulus: ; ; Denominator result: ; Similarity calculation: ; The results show that the semantic similarity scores of the three key semantic tags are ,This high similarity indicates that the semantic label contents of the two data sources are highly consistent, ,providing a quantitative basis for the subsequent correlation analysis.

[0034] Based on the semantic similarity score, the correlation between different data sources is evaluated, and the key related contents are marked to obtain semantic correlation data; The semantic tags with higher weights in the tag list are screened and matched, and all tags in the two lists are sorted in descending order according to the similarity scores. For tag pairs with scores higher than the threshold, the reliability of their matching in the context is recalculated, and the similarity scores are further adjusted by redistributing the weights. Finally, the tag pairs with higher similarity are marked, and all marked tag pairs are organized into semantic association data.

[0035] See also Figure 7 , the specific steps for obtaining semantic index configuration are: Based on semantic relevance, the index infrastructure is established, including keywords and their connection strength, to obtain a preliminary retrieval framework; A set of highly correlated keywords is extracted from the previous semantic similarity analysis results, and the keywords are analyzed with the corresponding semantic relationships and connection strengths. The co-occurrence frequency and contextual similarity of each pair of keywords are calculated to determine the strength of their connection. Based on this, a preliminary keyword network structure is constructed. Each node in the network represents a keyword, and the edge weight represents the connection strength, generating a preliminary retrieval framework.

[0036] Based on the preliminary retrieval framework, the information retrieval process is optimized, and the information deviation in the index is dynamically adjusted and corrected through query testing and user feedback to obtain the semantic index configuration; The retrieval capability of the keyword network is tested through actual query operations, the degree of match between each retrieval result and the expected result is analyzed, the keyword connection weight and node priority of the matching difference part are dynamically adjusted, and the information deviation is corrected in combination with user feedback. During the optimization process, the retrieval effect of each node is evaluated separately, and the consistency and accuracy of the results are improved by adjusting the upper and lower limits of the node connection weight to obtain the semantic index configuration.

[0037] See also Figure 8 , the specific steps for obtaining search optimization summaries are: Based on the semantic index configuration, keyword analysis is performed, and targeted search queries are executed according to the user's historical behavior and preference data to obtain a preliminary search result set; Parse the keywords and query content entered by the user, extract core keywords and generate a keyword set by analyzing the semantic relationship, word frequency distribution and context logic of the query, combine the user's historical behavior data, including browsing history, click frequency, page dwell time and user behavior preference analysis, and expand keywords based on the relevance of the query content. The expansion process is based on the logic of synonyms, near-synonyms and hyponyms, and matches the expanded phrases by calling the language database to generate an expanded keyword list, which is input into the search engine interface to generate a preliminary search result set.

[0038] Based on the preliminary search result set, the formula is used: ; Calculate the score for each search result , and automatically process the content summary to obtain an optimized content summary, where Indicates A visibility score for search results based on the relevance of the result content, click-through rate, and user interaction data. Indicates that for The user's personalized weight for each search result is used to measure the user's potential interest in the result based on the user's past search behavior and preference settings. is the number of search results; Extract the core indicators of each search result, including visibility score, user preference weight and content quality score. The visibility score is evaluated by calculating the user click-through rate, dwell time and interaction frequency. The click-through rate is calculated by counting the ratio of user clicks to total impressions, and the dwell time is calculated by the ratio of total page dwell time to number of visits. The interaction frequency counts the frequency of user mouse scrolling and clicking behaviors on the result page. The user preference weight is calculated based on the matching degree and relevance score of historical search behaviors. The content quality score is quantitatively scored based on the keyword matching in the search result text. The integrated indicators are used to comprehensively calculate each result. After calculating the score of each search result and sorting it, an optimized content summary is obtained. If is 1, the click rate of the first search result is 0.12, the average dwell time is 4 minutes, and the interaction frequency is 0.8. Comprehensive calculation: ; User preference weight According to the quantification of the matching degree and similarity of historical search behaviors, if the matching degree is 0.6 and the similarity is 0.7, we can get: ; Calculate the score of the first search result according to the formula: ; The result shows that the comprehensive score of the first search result is 0.42, which is used as the basis for subsequent sorting and optimization.

[0039] Sorting and display optimization of optimized content summaries, adjusting the display method and order according to user preferences, and obtaining search optimized summaries; Extract the core description text of each result in the search results, generate a content summary by counting the high-frequency words and key phrases in the text, call the record data of user browsing preferences, including the statistical results of the click-through rate of different types of summaries, prioritize the summaries according to the coverage of high-frequency key phrases, and at the same time, comprehensively consider the display and click preference types in user behavior, adjust the display order of the summaries, give priority to displaying the summary content that matches the user behavior pattern, complete the optimization and display adjustment of the content summary, and obtain the search optimized summary.

[0040] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A personalized information search system for different types of handicapped groups, characterized by: The system comprises: The speech-to-text conversion module analyzes the spectral characteristics of the audio signal based on the voice command issued by the user, calculates the energy peak and resonance peak position of the signal, identifies the energy distribution of multiple frequency bands, and extracts the key phoneme sequence, matches it with the grammatical rules, and then converts it into structured text to obtain the speech data analysis result; The visual text synthesis module collects the image data uploaded by users, identifies the objects and scenes in the image, extracts the text information in the image, analyzes the color, shape and size parameters, and converts the identified visual information into text description to obtain the visual information encoding result; The semantic fusion and retrieval module performs text information matching and sorting based on the speech data analysis results and visual information encoding results, analyzes the semantic relevance of different data sources, optimizes the information retrieval process, and corrects information deviations to obtain a semantic index configuration; The user-targeted recommendation module analyzes the user's search behavior and preference data based on the semantic index configuration, executes search queries, filters the search results of the queries, extracts information related to user needs, and automatically processes content summaries, and then sorts and optimizes the display of the summary information to obtain a search optimized summary.

2. The personalized information search system for different types of handicap groups according to claim 1 is characterized in that: The steps of identifying the energy distribution of the multiple frequency bands are specifically as follows: Based on the voice commands issued by the user, noise reduction and standardization are performed, and the time domain signal is converted into a frequency domain signal using fast Fourier transform to obtain spectrum data; Based on the spectrum data, the formula is adopted: ； Measure the energy density within the frequency band to obtain the energy peak and resonance peak position of each frequency band, where: is at frequency The energy value below is the sample size, is at frequency and sample index The complex output of the FFT result is yes The module length; Based on the energy peak and resonance peak position of each frequency band, the audio spectrum is analyzed to divide it into low frequency, medium frequency and high frequency bands, and the total energy of each frequency band is calculated to obtain the energy distribution of multiple frequency bands.

3. The personalized information search system for different types of handicap groups according to claim 1 is characterized in that: The steps for obtaining the speech data analysis result are specifically as follows: Based on the energy distribution of the multiple frequency bands and in combination with the spectrum data features, phoneme recognition is performed, and a preliminary phoneme recognition result is obtained by comparing the frequency of the energy peak in the frequency band with the spectrum features of the known phonemes; Based on the preliminary phoneme recognition results, the formula is adopted: ； Re-evaluate the matching degree between the phoneme sequence and the grammatical rules, verify the correctness of the word order, and adjust or replace the unmatched phonemes to obtain the grammatical matching result, where: is the posterior probability, indicating that when the phoneme sequence is detected Under the condition of The probability of occurrence, is the likelihood probability, expressed in phonemes Under the condition of occurrence, the phoneme sequence is detected The probability of It is a phoneme The prior probability of is a phoneme sequence The marginal probability of Based on the grammatical matching result, the phoneme sequence is converted into a structured text output, including translating the phonemes into corresponding texts, and applying linguistic rules to optimize the semantic level to obtain the speech data analysis result.

4. The personalized information search system for different types of handicap groups according to claim 1 is characterized in that: The analysis steps of the color, shape and size parameters are specifically as follows: Collect the image data uploaded by users, process the uploaded images, including color balance, brightness adjustment and contrast optimization, and obtain optimized image data; Based on the optimized image data, the color, shape and size parameters of multiple objects in the image are calculated using the formula: ； Get the feature score of each object, where is the average color value of the object, is the shape complexity, is the size parameter, is the color weight coefficient, is the shape weight coefficient, is the size weight coefficient; Based on the feature score of each object, the objects are classified and sorted, and the objects with key scores are processed first to obtain the color, shape and size parameter analysis results.

5. The personalized information search system for different types of handicap groups according to claim 1 is characterized in that: The steps of obtaining the visual information encoding result are specifically as follows: Based on the color, shape and size parameters, matching is performed with the visual information template to determine the classification label of each object and obtain a preliminary labeling result; Based on the preliminary labeling result, it is converted into a text description, the fluency and accuracy of the text description are adjusted, and the text description is encoded into a data format to obtain a visual information encoding result.

6. The personalized information search system for different types of handicap groups according to claim 1 is characterized in that: The analysis steps of the semantic relevance of the difference data sources are specifically as follows: Based on the speech data analysis results and the visual information encoding results, extract key semantic tags for each data source to obtain two independent key semantic tag lists; The two independent key semantic tag lists are cross-compared using the formula: ； Compute the normalized cosine similarity between label vectors , and get the semantic similarity score, where Representative The weight of key semantic tags, Representatives and The semantic score of the matching tags, Represents the total number of key semantic tags; Based on the semantic similarity score, the relevance between the different data sources is evaluated, and the key related contents are marked to obtain semantic relevance data.

7. The personalized information search system for different types of handicapped groups according to claim 1 is characterized in that: The steps for obtaining the semantic index configuration are specifically as follows: Based on the semantic relevance, an index infrastructure is established, including keywords and their connection strengths, to obtain a preliminary retrieval framework; Based on the preliminary retrieval framework, the information retrieval process is optimized, and the information deviation in the index is dynamically adjusted and corrected through query testing and user feedback to obtain a semantic index configuration.

8. The personalized information search system for different types of handicapped groups according to claim 1 is characterized in that: The steps for obtaining the search optimization summary are specifically as follows: Based on the semantic index configuration, keyword analysis is performed, and according to the user's historical behavior and preference data, a targeted search query is executed to obtain a preliminary search result set; Based on the preliminary search result set, the formula is used: ； Calculate the score for each search result , and automatically process the content summary to obtain an optimized content summary, where Indicates Visibility score for search results, Indicates that for User personalized weights of search results, is the number of search results; The optimized content summaries are sorted and displayed in an optimized manner, and the display method and order are adjusted according to the user's preference to obtain a search optimized summary.

Images