Data search method, device, and storage medium

Abstract

The application discloses a data search method and device and a storage medium, and relates to the technical field of information retrieval, and the method comprises the following steps: acquiring search content of a user; recalling data assets corresponding to the search content based on a preset search strategy to obtain a plurality of recall results, wherein the search strategy is obtained by combining a plurality of dimensions. The application can meet the user demand in a specific use scenario.

Description

Technical Field

[0001] This invention relates to the field of information retrieval technology, and in particular to a data search method, device, and storage medium. Background Technology

[0002] Data asset search provides a convenient way for humans to access information and solve various problems. Through data asset search, people can access a wealth of data resources, including internal and external corporate data, academic research papers and datasets, market research reports and data, publicly available government data, and more. These data resources can help people gain profound insights, effectively supporting their work in decision-making, research and analysis, market research, and other areas across various fields.

[0003] Currently, data asset search scenarios typically involve matching user input with data asset names to retrieve results, which are then returned to the user. This strategy usually only considers the single dimension of text matching between the search content and the candidate set of data assets. For general search scenarios, this strategy can generally meet users' search needs. However, in specific use cases, such as company employees searching for data assets in the company's data warehouse, considering only the text matching dimension is insufficient, and the above search strategies are difficult to meet user needs. Summary of the Invention

[0004] The main objective of this invention is to provide a data search method, device, and storage medium, aiming to solve the problem that current data asset search methods only consider the text matching dimension and cannot meet the user needs in specific use cases.

[0005] To achieve the above objectives, the present invention provides a data search method, the method comprising:

[0006] Obtain the user's search content;

[0007] Based on a preset search strategy, data assets corresponding to the search content are retrieved to obtain several retrieval results. The search strategy is set by combining several dimensions.

[0008] Optionally, the step of recalling data assets corresponding to the search content based on a preset search strategy to obtain several recall results includes:

[0009] The data asset score values ​​of the several recall results are calculated based on a preset weight allocation strategy;

[0010] Based on the data asset scores of the recall results, the recall results are sorted, and the sorted recall results are returned to the user.

[0011] Optionally, the step of obtaining the user's search content is followed by:

[0012] The search results are preprocessed to obtain processed text;

[0013] The step of recalling data assets corresponding to the search content based on a preset search strategy to obtain several recall results, wherein the search strategy is obtained by combining several dimensions, includes:

[0014] Based on the preset search strategy, the data assets corresponding to the processed text are recalled to obtain the plurality of recall results. The search strategy is set by combining the plurality of dimensions.

[0015] Optionally, the step of preprocessing the search content to obtain processed text includes:

[0016] Determine whether the search content contains spaces;

[0017] If the search content does not contain spaces, then the search content is segmented into words to obtain the processed text;

[0018] If the search content contains the space, then the search content is divided into multiple keywords based on the space;

[0019] The keywords are segmented into words, and the keywords and the segmented keywords are used as the processed text.

[0020] Optionally, the step of recalling data assets corresponding to the processed text based on a preset search strategy to obtain several recall results includes:

[0021] The processed text is matched with the names of each data asset to obtain the first recall result;

[0022] And / or, determine whether the processed text contains Chinese names. If the processed text contains Chinese names, perform semantic similarity retrieval on the processed text and each of the data assets to obtain a second recall result.

[0023] And / or, recall the browsing history assets of the user's department colleagues to obtain a third recall result;

[0024] And / or, recall data assets that are related to the user's historical browsing assets to obtain a fourth recall result;

[0025] And / or, recall data assets whose pageviews exceed a preset pageview threshold and / or whose authorizations exceed a preset authorization threshold, to obtain the fifth recall result.

[0026] Optionally, the step of performing semantic similarity retrieval on the processed text and each of the data assets to obtain the second recall result includes:

[0027] Convert the processed text into a processed text vector;

[0028] Convert the names of each data asset into name vectors for each data asset.

[0029] Calculate the cosine similarity between the processed text vector and the name vector of each data asset to obtain the first similarity of each data asset;

[0030] The similarity between the processed text vector and each field in each of the data assets is calculated.

[0031] The maximum similarity between the processed text vector and each field in each of the data assets is taken as the second similarity of each of the data assets;

[0032] The first similarity and the second similarity of the data assets are compared, and the larger value is taken as the final similarity of the data assets.

[0033] The final similarity scores of each of the data assets are sorted to obtain a sorting result;

[0034] The sorting results are selected and recalled based on a preset sorting result selection strategy to obtain the second recall result.

[0035] Optionally, the step of recalling the historical browsing assets of the user's department colleagues to obtain a third recall result includes:

[0036] Obtain the user's employee ID;

[0037] Based on the user's employee ID, the employee IDs of the user's department colleagues can be retrieved;

[0038] Based on the employee ID of the colleague in the department, the historical browsing assets of the colleague in the department are obtained, and the third recall result is obtained.

[0039] Optionally, the step of recalling data assets related to the user's historical browsing assets to obtain a fourth recall result includes:

[0040] Obtain the user's employee ID;

[0041] The user's historical browsing assets can be retrieved based on the user's employee ID;

[0042] Based on the user's historical browsing assets, find the upstream and downstream assets related to the user's historical browsing assets;

[0043] The upstream and downstream assets related to the blood relationship are recalled, resulting in the fourth recall result.

[0044] This invention also proposes a device, which includes a memory, a processor, and a data search program stored in the memory and executable on the processor. When the data search program is executed by the processor, it implements the data search method described above.

[0045] This invention also proposes a computer-readable storage medium storing a data search program, which, when executed by a processor, implements the data search method described above.

[0046] The data search method, device, and storage medium proposed in this invention acquire user search content; based on a preset search strategy, retrieve data assets corresponding to the search content to obtain several retrieved results. The search strategy is obtained by combining several dimensions. This invention retrieves data assets corresponding to the search content based on a preset search strategy combining several dimensions, obtaining retrieved results. By optimizing the retrieved content by combining multiple dimensions, the retrieved results can better match user intent and meet user needs in specific usage scenarios. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the functional modules of the device to which the data search device of the present invention belongs;

[0048] Figure 2 This is a flowchart illustrating an exemplary embodiment of the data search method of the present invention;

[0049] Figure 3 This is a flowchart illustrating another exemplary embodiment of the data search method of the present invention;

[0050] Figure 4 This is a flowchart illustrating another exemplary embodiment of the data search method of the present invention;

[0051] Figure 5 This is a flowchart illustrating another exemplary embodiment of the data search method of the present invention;

[0052] Figure 6 This is a flowchart illustrating another exemplary embodiment of the data search method of the present invention;

[0053] Figure 7 This is a flowchart illustrating another exemplary embodiment of the data search method of the present invention;

[0054] Figure 8 This is a flowchart illustrating another exemplary embodiment of the data search method of the present invention;

[0055] Figure 9 This is a flowchart illustrating another exemplary embodiment of the data search method of the present invention;

[0056] Figure 10 This is a schematic diagram illustrating the overall process of data search in an embodiment of the present invention.

[0057] 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

[0058] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0059] The main solution of this invention is: to obtain the user's search content; to recall data assets corresponding to the search content based on a preset search strategy, resulting in several recall results. The search strategy is obtained by combining several dimensions. This invention recalls data assets corresponding to the search content based on a preset search strategy combining several dimensions, obtaining recall results. By optimizing the recalled content by combining multiple dimensions, the recall results can better match the user's intent and meet the user's needs in specific usage scenarios.

[0060] Technical terms involved in the embodiments of this invention:

[0061] Data assets: Data assets refer to the data resources that an organization or enterprise owns, creates, stores, and manages. They are an important business asset and competitive advantage.

[0062] Recall: Recall refers to how many texts in a set of relevant texts match a user's search query. Simply put, recall is the proportion of retrieved texts relevant to the user's search query out of all relevant texts.

[0063] Semantic similarity retrieval: Semantic similarity retrieval refers to matching texts based on their semantic meaning. In this case, natural language processing techniques, such as word vector models (e.g., Word2Vec, BERT), are typically used to represent the semantic information of the text, and then the similarity between texts is calculated. Furthermore, semantic similarity retrieval considers both the semantic meaning of words and contextual information.

[0064] Cosine similarity: Cosine similarity is a method for measuring the degree of similarity between two vectors. It measures the correlation between them by calculating the cosine of the angle between the two vectors in a multidimensional space. It can be used to compare the similarity between any two vectors and is commonly applied in fields such as information retrieval, data mining, recommender systems, and natural language processing.

[0065] Word vector averaging to obtain sentence vectors: This is a simple and common method for text vector representation. In this method, the word vectors of each word in a sentence are summed, and the average value is taken as the representation vector of the entire sentence.

[0066] Lineage: In data management and use, lineage usually refers to the association or connection between different data assets. These associations or connections can be because they share the same data source, are used together in the same project or task, are related through operations such as transformation or processing, or are interdependent in some process.

[0067] This invention considers the following: Currently, data asset search scenarios typically involve matching user input with data asset names to retrieve results, which are then returned to the user. This process also preprocesses the user input and data asset names, such as word segmentation and removal of pause words, and calculates the similarity between the search content and candidate assets according to certain rules. These strategies usually only consider the single dimension of text matching between the search content and the candidate data asset set. For general search scenarios, these strategies can basically meet users' search needs. However, in specific use cases, such as company employees searching for data assets in the company's data warehouse, considering only the text matching dimension is insufficient, and the above search strategies are difficult to meet user needs.

[0068] Therefore, this invention proposes a solution that retrieves data assets corresponding to search content based on a preset search strategy that combines several dimensions, and obtains retrieval results. By optimizing the retrieved content by combining multiple dimensions, the retrieval results can better match user intent and meet user needs in specific usage scenarios.

[0069] Specifically, refer to Figure 1 , Figure 1 This is a functional module diagram of the device to which the data search device of the present invention belongs. The data search device can be a data processing device independent of the device itself, which can be carried on the device in hardware or software form. The device can be a smart mobile terminal with data processing capabilities, such as a mobile phone or tablet computer, or it can be a fixed device or server with data processing capabilities.

[0070] In this embodiment, the data search device includes at least an output module 110, a processor 120, a memory 130, and a communication module 140.

[0071] The memory 130 stores the operating system and data search program; the output module 110 may be a display screen, etc. The communication module 140 may include a WIFI module and a Bluetooth module, etc., and communicates with external devices or servers through the communication module 140.

[0072] When the data search program in memory 130 is executed by the processor, it performs the following steps:

[0073] Obtain the user's search content;

[0074] Based on a preset search strategy, data assets corresponding to the search content are retrieved to obtain several retrieval results. The search strategy is set by combining several dimensions.

[0075] Furthermore, when the data search program in memory 130 is executed by the processor, it also performs the following steps:

[0076] The data asset score values ​​of the several recall results are calculated based on a preset weight allocation strategy;

[0077] Based on the data asset scores of the recall results, the recall results are sorted, and the sorted recall results are returned to the user.

[0078] Furthermore, when the data search program in memory 130 is executed by the processor, it also performs the following steps:

[0079] The search results are preprocessed to obtain processed text;

[0080] Based on the preset search strategy, the data assets corresponding to the processed text are recalled to obtain the plurality of recall results. The search strategy is set by combining the plurality of dimensions.

[0081] Furthermore, when the data search program in memory 130 is executed by the processor, it also performs the following steps:

[0082] Determine whether the search content contains spaces;

[0083] If the search content does not contain spaces, then the search content is segmented into words to obtain the processed text;

[0084] If the search content contains the space, then the search content is divided into multiple keywords based on the space;

[0085] The keywords are segmented into words, and the keywords and the segmented keywords are used as the processed text.

[0086] Furthermore, when the data search program in memory 130 is executed by the processor, it also performs the following steps:

[0087] The processed text is matched with the names of each data asset to obtain the first recall result;

[0088] And / or, determine whether the processed text contains Chinese names. If the processed text contains Chinese names, perform semantic similarity retrieval on the processed text and each of the data assets to obtain a second recall result.

[0089] And / or, recall the browsing history assets of the user's department colleagues to obtain a third recall result;

[0090] And / or, recall data assets that are related to the user's historical browsing assets to obtain a fourth recall result;

[0091] And / or, recall data assets whose pageviews exceed a preset pageview threshold and / or whose authorizations exceed a preset authorization threshold, to obtain the fifth recall result.

[0092] Furthermore, when the data search program in memory 130 is executed by the processor, it also performs the following steps:

[0093] Convert the processed text into a processed text vector;

[0094] Convert the names of each data asset into name vectors for each data asset.

[0095] Calculate the cosine similarity between the processed text vector and the name vector of each data asset to obtain the first similarity of each data asset;

[0096] The similarity between the processed text vector and each field in each of the data assets is calculated.

[0097] The maximum similarity between the processed text vector and each field in each of the data assets is taken as the second similarity of each of the data assets;

[0098] The first similarity and the second similarity of the data assets are compared, and the larger value is taken as the final similarity of the data assets.

[0099] The final similarity scores of each of the data assets are sorted to obtain a sorting result;

[0100] The sorting results are selected and recalled based on a preset sorting result selection strategy to obtain the second recall result.

[0101] Furthermore, when the data search program in memory 130 is executed by the processor, it also performs the following steps:

[0102] Obtain the user's employee ID;

[0103] Based on the user's employee ID, the employee IDs of the user's department colleagues can be retrieved;

[0104] Based on the employee ID of the colleague in the department, the historical browsing assets of the colleague in the department are obtained, and the third recall result is obtained.

[0105] Furthermore, when the data search program in memory 130 is executed by the processor, it also performs the following steps:

[0106] Obtain the user's employee ID;

[0107] The user's historical browsing assets can be retrieved based on the user's employee ID;

[0108] Based on the user's historical browsing assets, find the upstream and downstream assets related to the user's historical browsing assets;

[0109] The upstream and downstream assets related to the blood relationship are recalled, resulting in the fourth recall result.

[0110] This embodiment, through the above-described scheme, specifically obtains the user's search content; based on a preset search strategy, it retrieves data assets corresponding to the search content, obtaining several retrieved results. The search strategy is obtained by combining several dimensions. This embodiment of the invention retrieves data assets corresponding to the search content based on a preset search strategy combined with several dimensions, obtaining retrieved results. By optimizing the retrieved content by combining multiple dimensions, the retrieved results can be made more in line with user intent and meet user needs in specific usage scenarios.

[0111] Based on, but not limited to, the above-described device architecture, embodiments of the method of the present invention are proposed.

[0112] The execution subject of the method in this embodiment can be a data search device, which can be a device that is independent of the device and capable of data processing. It can be carried on the device in the form of hardware or software.

[0113] Reference Figure 2 , Figure 2 This is a flowchart illustrating an exemplary embodiment of the data search method of the present invention. The data search method includes:

[0114] Step S10: Obtain the user's search content.

[0115] In one implementation method, users can enter search terms in the input box on the interactive page, and then the data search method of the present invention obtains the user's search terms.

[0116] Step S30: Based on a preset search strategy, retrieve the data assets corresponding to the search content to obtain several retrieval results. The search strategy is set by combining several dimensions.

[0117] Specifically, for the search content entered by the user, the data assets corresponding to the search content are searched based on a preset search strategy, and then the data assets corresponding to the search content are retrieved to obtain several retrieval results.

[0118] Specifically, as one implementation method, the following are some search strategies:

[0119] 1. Time Range Strategy: This strategy filters assets in the data assets that meet the time criteria based on the time range specified by the user, such as filtering by creation time or publication time.

[0120] 2. Location-Based Strategy: This strategy uses location information to perform location matching within data assets, finding assets related to the user's specified location. This can be used for searching and filtering based on location, such as searching for assets in or near a specific city.

[0121] 3. Hierarchical Structure Strategy: This strategy searches and retrieves assets hierarchically based on the user-selected hierarchical relationship within a data asset structure. For example, it can search by hierarchical relationships within an organizational structure or search from the overall structure to specific sub-structures.

[0122] 4. Tag Filtering Strategy: This strategy matches data assets with predefined tags or attributes to filter out assets that meet user requirements. This can be used for classifying, labeling, and filtering data assets.

[0123] 5. Filtering Strategy: This strategy filters the recall results based on specified filtering conditions to provide more accurate search results.

[0124] 6. Text matching strategy: Retrieval is achieved by matching the user's search text with the data asset name after segmenting the input text according to certain rules.

[0125] 7. Semantic similarity strategy: If the search content contains Chinese names, semantic similarity retrieval will be performed.

[0126] 8. Collaborative filtering strategy: Recall the browsing history of users' colleagues.

[0127] 9. Lineage Recommendation Strategy: Recall upstream and downstream assets that are related to the user's historical browsing assets.

[0128] 10. Popular Recommendation Strategy: Recall based on data asset pageviews, visits, and authorizations.

[0129] It should be noted that the preset search strategies are not limited to the few listed above, and can also be set according to business needs and user habits.

[0130] One or more search strategies can be selected to retrieve data assets corresponding to the search content, based on business needs and actual circumstances.

[0131] Each search strategy corresponds to one recall result.

[0132] Furthermore, as one implementation method, after obtaining several recall results, the data asset score of the several recall results can be calculated based on a preset weight allocation strategy.

[0133] Then, based on the data asset scores of the recall results, the recall results are sorted, and the sorted recall results are returned to the user.

[0134] This embodiment, through the above-described scheme, obtains the user's search content; based on a preset search strategy, it retrieves the data assets corresponding to the search content, obtaining several retrieved results. The search strategy is obtained by combining several dimensions. This embodiment of the invention retrieves the data assets corresponding to the search content based on a preset search strategy combined with several dimensions, obtaining retrieved results. By optimizing the retrieved content by combining multiple dimensions, the retrieved results can better match the user's intent and meet the user's needs in specific usage scenarios.

[0135] Reference Figure 3 , Figure 3 This is a flowchart illustrating another exemplary embodiment of the data search method of the present invention.

[0136] Based on the above Figure 2 In the embodiment shown, step S30, after retrieving data assets corresponding to the search content based on a preset search strategy and obtaining several retrieval results, includes:

[0137] Step S40: Calculate the data asset score value of the several recall results based on the preset weight allocation strategy.

[0138] Different weights are assigned to different search strategies for different dimensions, and data asset scores are calculated for the different recall results obtained by different search strategies for different dimensions.

[0139] Then, the different recall results are sorted according to the data asset score, for example, from largest to smallest, and returned to the user.

[0140] Optionally, the weights of different search strategies can be flexibly adjusted based on user feedback. This allows users to prioritize browsing results that interest them, meeting personalized search needs and optimizing the user experience. For example, if users prioritize searches that directly match the name of a data asset, the weight of the text matching strategy can be increased; if users prioritize the semantic similarity between the search content and the name of the data asset, the weight of the semantic similarity strategy can be increased.

[0141] Specifically, assuming there are 5 search strategies, corresponding to 5 recall results, the sum of the data asset scores of the 5 recall results can be represented as follows:

[0142] score=d1+d2+d3+d4+d5

[0143] Step S50: Sort the recall results based on the data asset scores of the recall results, and return the sorted recall results to the user.

[0144] Among them, all the different recall results obtained based on different search strategies will be displayed to the user. Different weights are assigned to different search strategies for different dimensions, so that the display order of different recall results is different.

[0145] This embodiment, through the above-described scheme, obtains the user's search content; recalls data assets corresponding to the search content based on a preset search strategy, obtaining several recall results. The search strategy is set in combination with several dimensions. Specifically, a data asset score value is calculated for the several recall results based on a preset weight allocation strategy; the several recall results are sorted based on the data asset score values, and the sorted several recall results are returned to the user.

[0146] This invention recalls data assets corresponding to search content based on a preset search strategy that combines several dimensions, yielding recall results. By optimizing the recalled content across multiple dimensions, the results better align with user intent and meet user needs in specific usage scenarios. Specifically, this invention calculates data asset scores for several recall results based on a preset weighting strategy, and then sorts these results based on these scores. The weighting strategy can be flexibly adjusted based on user feedback to adjust the weights of recall results obtained from different dimensions. This allows users to prioritize recall results that interest them, satisfying personalized search needs and optimizing the user experience.

[0147] Refer to Figure 4 , Figure 4 which is a schematic flowchart of another exemplary embodiment of the data search method of the present invention.

[0148] Based on the above Figure 2 shown embodiment, after the step S10 of obtaining the search content of the user, it includes:

[0149] Step S20, preprocess the search content to obtain a processed text.

[0150] Among them, preprocessing the search content can reduce interference and noise, thereby improving the accuracy of the recall result; and by preprocessing the search content, the search content can be simplified, redundant information and complex structures can be removed, thereby reducing the complexity of the search and improving the search efficiency.

[0151] Specifically, as an implementation manner, the following lists several ways to preprocess the search content:

[0152] 1. Word segmentation: According to the user's search habit, when the search content contains multiple keywords, they are generally separated by spaces, so the space is used as an important separation mark. Determine whether the search content contains spaces. If the search content contains spaces, the search content is divided into multiple keywords based on the spaces, and then the keywords are further segmented. For example, the keyword "fund fixed income" is segmented into "fund" and "fixed income", and the keywords and the keywords after segmentation are used as the processed text; if the search content does not contain spaces, the search content is segmented to obtain the processed text.

[0153] 2. Stop word removal: Stop words refer to common words that have no actual meaning in the search and appear frequently, such as "of", "is", "in", etc. Removing these stop words can reduce interference and improve the quality of the search results.

[0154] 3. Abbreviation expansion: If the search content contains abbreviations or shorthand forms, they can be expanded into complete words to increase the accuracy of matching.

[0155] 4. Special character deletion: Delete special characters in the search content, such as punctuation marks, special symbols or emojis, to avoid interfering with the search results.

[0156] 5. Standardization and normalization: For searches in special fields or specific requirements, specific standardization and normalization processing can be performed, such as uniformly formatting dates, times, units, etc.

[0157] It should be noted that the methods for preprocessing search content are not limited to those listed above. Any search content processing method that helps to clean and standardize search content and improve matching results and search accuracy can be considered a method for preprocessing search content.

[0158] One or more preprocessing methods can be selected to preprocess the search content.

[0159] Reference Figure 5 , Figure 5 This is a flowchart illustrating another exemplary embodiment of the data search method of the present invention.

[0160] Based on the above Figure 2 In the embodiment shown, step S30 involves recalling data assets corresponding to the search content based on a preset search strategy, resulting in several recall results. The search strategy is derived by combining several dimensions, including:

[0161] Step S31: Based on the preset search strategy, recall the data assets corresponding to the processed text to obtain the plurality of recall results. The search strategy is set by combining the plurality of dimensions.

[0162] Specifically, for the preprocessed text, the data assets corresponding to the preprocessed text are searched based on a preset search strategy, and then the data assets corresponding to the preprocessed text are recalled to obtain several recall results.

[0163] This embodiment, through the above-described scheme, obtains the user's search content; recalls data assets corresponding to the search content based on a preset search strategy, obtaining several recall results. The search strategy is set in combination with several dimensions. Specifically, the search content is preprocessed to obtain processed text; the data assets corresponding to the processed text are recalled based on the preset search strategy, obtaining the several recall results. The search strategy is set in combination with the several dimensions.

[0164] This invention, based on a preset search strategy combining several dimensions, retrieves data assets corresponding to search content, yielding retrieval results. By optimizing the retrieved content across multiple dimensions, the results better align with user intent and meet user needs in specific usage scenarios. Specifically, this invention preprocesses the search content to reduce interference and noise, thereby improving the accuracy of the retrieval results. Furthermore, preprocessing simplifies the search content, removing redundant information and complex structures, thus reducing search complexity and improving search efficiency.

[0165] Reference Figure 6 , Figure 6 This is a flowchart illustrating another exemplary embodiment of the data search method of the present invention.

[0166] Based on the above Figure 4 , 5 In the embodiment shown, step S20, which preprocesses the search content to obtain processed text, includes:

[0167] Step S21: Determine whether the search content contains spaces.

[0168] In particular, based on user search habits, when the search content contains multiple keywords, they are usually separated by spaces, so spaces are used as an important separator.

[0169] Step S22: If the search content does not contain spaces, then the search content is segmented to obtain the processed text.

[0170] If the search content does not contain spaces, then the search content is directly segmented into words.

[0171] Step S23: If the search content contains the space, then the search content is divided into multiple keywords based on the space.

[0172] Step S24: Perform word segmentation on the keywords, and use the keywords and the segmented keywords as the processed text.

[0173] The search results entered by the user are divided into spaces, and the resulting content can be further segmented into words, such as segmenting "fund fixed income" into two words: "fund" and "fixed income".

[0174] By segmenting search results into words, long strings can be broken down into more representative terms, allowing for more accurate matching of relevant data assets and improving the accuracy of search results. Furthermore, for search results containing spaces, segmenting based on spaces and then segmenting the resulting keywords transforms them into more easily understood and processed text, thus providing more flexible search results.

[0175] Furthermore, after obtaining the processed text, it can be matched with the data asset name, and a recall can be performed when the data asset name contains any word from the processed text.

[0176] This embodiment, through the above-described scheme, obtains the user's search content; recalls data assets corresponding to the search content based on a preset search strategy, obtaining several recall results. The search strategy is set in combination with several dimensions. Specifically, the search content is preprocessed to obtain processed text; the data assets corresponding to the processed text are recalled based on the preset search strategy, obtaining the several recall results. The search strategy is set in combination with the several dimensions. Specifically, it is determined whether the search content contains spaces; if the search content does not contain spaces, the search content is segmented into words to obtain the processed text; if the search content contains spaces, the search content is divided into multiple keywords based on the spaces; the keywords are segmented into words, and the keywords and the segmented keywords are used as the processed text.

[0177] This invention, based on a pre-defined search strategy combining several dimensions, retrieves data assets corresponding to search content, yielding retrieval results. By optimizing the retrieved content across multiple dimensions, the results better align with user intent and meet user needs in specific usage scenarios. Specifically, preprocessing the search content reduces interference and noise, improving the accuracy of retrieval results. Furthermore, preprocessing simplifies the search content, removing redundant information and complex structures, thus reducing search complexity and improving efficiency. By segmenting the search content into more representative words, long strings can be divided into more accurate matching of relevant data assets, enhancing search result accuracy. Moreover, for search content containing spaces, segmentation based on spaces and subsequent keyword segmentation transforms the content into more easily understood and processed text, providing more flexible search results.

[0178] Based on the above Figure 6 In the embodiment shown, step S31 involves recalling data assets corresponding to the processed text based on a preset search strategy, resulting in several recall results, including:

[0179] Step S311: Match the processed text with the names of each data asset to obtain the first recall result.

[0180] Specifically, as one implementation method, when determining whether the search content contains spaces, the processed text can be matched with the names of various data assets to obtain the first recall result.

[0181] If the search content does not contain spaces, the search content is segmented to obtain the processed text. The processed text is then matched with the names of each data asset. If the name of a data asset contains any keyword in the processed text, that data asset is recalled.

[0182] If the search content contains the space, the search content is divided into multiple keywords based on the space. Then, the multiple keywords are matched with the names of each data asset. If the name of a data asset contains any keyword, the data asset is recalled.

[0183] Then, the keywords are segmented, and the keywords and the segmented keywords are used as the processed text. The processed text is then matched with the names of each data asset. If the name of a data asset contains any keyword in the processed text or any keyword after segmentation, the data asset is recalled.

[0184] Specifically, the following example illustrates how to match the processed text with the names of various data assets to obtain the first recall result:

[0185] First, the search content is split into multiple keywords by spaces. For each keyword, if the data asset name directly contains the keyword, the number of words after the keyword is segmented is calculated and used as indicator 1, denoted as k1.

[0186] If other information of the data asset (such as field names, descriptions, etc.) directly contains the keyword, calculate the number of words after the keyword is segmented, and use it as indicator 2, denoted as k2 (keywords that were previously calculated will not be calculated here).

[0187] If both indicators 1 and 2 above are zero, it means that the keyword does not appear directly in the asset information. Further word segmentation of the keyword is needed. After word segmentation, multiple words are obtained. The number of words contained in the data asset name is calculated as indicator 3, denoted as k3. The number of words contained in other information of the data asset is calculated as indicator 4, denoted as k4 (words calculated previously are not counted here). The weights of these four indicators decrease sequentially from top to bottom. The scores d1 for direct text matching in this dimension are calculated by summing them up. d1 can be represented as follows:

[0188] d1 = k1 + k2 + k3 + k4

[0189] Step S312, and / or, determine whether the processed text contains Chinese names. If the processed text contains Chinese names, perform semantic similarity retrieval on the processed text and each of the data assets to obtain a second recall result.

[0190] In one implementation method, the processed text is first converted into a processed text vector.

[0191] Then, the names of each data asset are converted into name vectors for each data asset.

[0192] Then, the cosine similarity between the processed text vector and the name vector of each data asset is calculated to obtain the first similarity of each data asset.

[0193] Then, the similarity between the processed text vector and each field in each of the data assets is calculated.

[0194] Then, the maximum similarity between the processed text vector and each field in each of the data assets is taken as the second similarity of each of the data assets.

[0195] Then, the first similarity and the second similarity of the data assets are compared, and the larger value is taken as the final similarity of the data assets.

[0196] Then, the final similarity of each data asset is sorted to obtain the sorting result.

[0197] Finally, the sorting results are selected and recalled based on a preset sorting result selection strategy to obtain the second recall result.

[0198] Step S313, and / or, recall the historical browsing assets of the user's department colleagues to obtain a third recall result.

[0199] Among them, the historical browsing assets of department colleagues are the data assets that department colleagues have browsed.

[0200] Among them, colleagues in the same department have strong business connections, and the user's target data assets are likely to have been used and viewed by other colleagues in the department as well. Recalling search results by reviewing the browsing history of colleagues in the department can also optimize the recall results to some extent.

[0201] In one implementation method, the user's employee ID is first obtained.

[0202] Then, based on the user's employee ID, the employee IDs of the user's department colleagues are retrieved.

[0203] Finally, based on the employee ID of the colleague in the department, the historical browsing assets of the colleague in the department are retrieved, and the third recall result is obtained.

[0204] Step S314, and / or, recall data assets that are related to the user's historical browsing assets to obtain a fourth recall result.

[0205] Among them, data assets that are related to the user's historical browsing assets are data assets that are associated with or connected to the user's historical browsing assets. These associations or connections may be because they share the same data source, are used together in the same project or task, are related to each other through operations such as transformation or processing, or are interdependent in some process.

[0206] The names of data assets are numerous and complex, and their naming is highly correlated with project names and personal naming habits, making it difficult for users to remember them completely. By combining the relevance of assets viewed in the user's history, the burden on the user's memory can be reduced, making it easier for them to find and revisit previously viewed related assets. Furthermore, some data assets are important and used by users in different projects, leading to repeated searches for the same data asset. Additionally, related data assets have strong connections; therefore, recalling assets viewed in the user's history and those with related connections can optimize the retrieval results.

[0207] In one implementation method, the user's employee ID is first obtained.

[0208] Then, the user's historical browsing assets are retrieved based on the user's employee ID.

[0209] Then, based on the user's historical browsing assets, the upstream and downstream assets related to the user's historical browsing assets are found.

[0210] Finally, the upstream and downstream assets related to the blood relationship are recalled, resulting in the fourth recall result.

[0211] Step S315, and / or, recall data assets whose pageviews exceed a preset pageview threshold and / or whose authorizations exceed a preset authorization threshold, to obtain the fifth recall result.

[0212] Specifically, the following example illustrates how the fifth recall result was obtained:

[0213] The higher the data asset's page views and authorization count, the more people search for and access that asset, and the more important it is often. Therefore, this part of the content is recalled. Here, v is the page views and a is the authorization count, which can be expressed by the following equation:

[0214] d5 = log(v + a + 1);

[0215] It should be noted that this embodiment does not limit the order in which the five search strategies are used to obtain the five recall results listed above.

[0216] Furthermore, by considering five dimensions—text matching, text similarity, browsing by colleagues within the same department, and users' historical search content—the coverage of retrieved results can be increased, improving the recall rate. This helps users find more relevant data assets. It also allows for the optimization of retrieved content, making search results more aligned with user intent.

[0217] This embodiment, through the above-described scheme, obtains the user's search content; recalls data assets corresponding to the search content based on a preset search strategy, obtaining several recall results. The search strategy is set in combination with several dimensions. Specifically, the search content is preprocessed to obtain processed text; the data assets corresponding to the processed text are recalled based on the preset search strategy, obtaining the several recall results. The search strategy is set in combination with the several dimensions. Specifically, it is determined whether the search content contains spaces; if the search content does not contain spaces, the search content is segmented into words to obtain the processed text; if the search content contains spaces, the search content is divided into multiple keywords based on the spaces; the keywords are segmented into words, and the keywords and the segmented keywords are used as the processed text. Specifically, the following steps are performed: matching the processed text with the names of each data asset to obtain a first recall result; and / or determining whether the processed text contains Chinese names; if the processed text contains Chinese names, performing semantic similarity retrieval on the processed text and each of the data assets to obtain a second recall result; and / or recalling the historical browsing assets of the user's department colleagues to obtain a third recall result; and / or recalling data assets related to the user's historical browsing assets to obtain a fourth recall result; and / or recalling data assets with browsing volume greater than a preset browsing volume threshold and / or authorization volume greater than a preset authorization volume threshold to obtain a fifth recall result.

[0218] This invention, based on a pre-defined search strategy combining several dimensions, retrieves data assets corresponding to search content, yielding retrieved results. By optimizing the retrieved content across multiple dimensions, the results better align with user intent and meet user needs in specific usage scenarios. Specifically, preprocessing the search content reduces interference and noise, improving the accuracy of the retrieved results. Furthermore, preprocessing simplifies the search content, removing redundant information and complex structures, thus reducing search complexity and improving efficiency. By segmenting the search content into more representative words, long strings can be divided into more precise matching of relevant data assets, improving search result accuracy. For search content containing spaces, segmentation based on spaces and subsequent keyword segmentation transforms the content into more easily understood and processed text, providing more flexible search results. Finally, by considering five dimensions—text matching, text similarity, browsing by colleagues within the same department, and user's historical search content—this invention increases the coverage of retrieved results and improves the recall rate, helping users find more relevant data assets. It can also optimize the retrieved content, making search results more aligned with user intent. Furthermore, the higher the number of page views and authorizations for a data asset, the more people search for and access that asset, indicating its importance. Therefore, retrieving this portion of data assets can optimize the retrieval results.

[0219] Reference Figure 7 , Figure 7 This is a flowchart illustrating another exemplary embodiment of the data search method of the present invention.

[0220] As one embodiment, step S312, and / or determining whether the processed text contains Chinese names, if the processed text contains Chinese names, then performing semantic similarity retrieval on the processed text and each of the data assets to obtain a second recall result includes:

[0221] Step S3121: Convert the processed text into a processed text vector.

[0222] Step S3122: Convert the names of each data asset into name vectors for each data asset.

[0223] For the Chinese names of each data asset, the name vector of each data asset can be calculated by averaging word vectors to obtain sentence vectors.

[0224] Step S3123: Calculate the cosine similarity between the processed text vector and the name vector of each data asset to obtain the first similarity of each data asset.

[0225] Step S3124: Calculate the similarity between the processed text vector and each field in each of the data assets.

[0226] Specifically, for Chinese fields in each data asset, all Chinese field names in the same data asset can be concatenated to obtain a concatenated word.

[0227] Then, by segmenting the concatenated words, multiple words representing field information are obtained, and each word is converted into a word vector.

[0228] Then, the cosine similarity between each single vector and the processed text vector is calculated.

[0229] Step S3125: The maximum similarity between the processed text vector and each field in each of the data assets is taken as the second similarity of each of the data assets.

[0230] Step S3126: Compare the first similarity and the second similarity of the data asset, and take the larger value as the final similarity of the data asset.

[0231] Step S3127: Sort the final similarity of each data asset to obtain the sorting result.

[0232] One implementation method is to sort the final similarity of each data asset from largest to smallest to obtain a sorting result.

[0233] Step S3128: Select and recall the sorting results based on a preset sorting result selection strategy to obtain the second recall result.

[0234] One approach is to sort the data assets by their final similarity scores from highest to lowest and select the k most similar data assets as the second recall result.

[0235] Specifically, the following example illustrates how to obtain a second recall result:

[0236] The names of data assets are segmented into words, and the average of the segmented words is calculated to obtain the data asset name vector w. L Let w be the word vector of the processed text. s The calculated similarity between the data asset name and the word vector of the processed text is denoted as d. 21 The calculated word vectors of the processed text and various fields of the data assets w k The maximum similarity is denoted as d. 22 Take d 21 and d 22The largest one is used as the final similarity score d2 for processing text and data assets. All candidate assets are sorted from largest to smallest based on their final similarity scores d2, and the top k are used for recall.

[0237] Among them, w L It can be represented as follows:

[0238] w L =1 / nsum(w i ), i = 1, ..., n;

[0239] Among them, w i This represents the vector of each word after word segmentation.

[0240] d 21 It can be represented as follows:

[0241] d 21 =cos(w L ,w s );

[0242] d 22 It can be represented as follows:

[0243] d 22 =max(cos(w) k ,w s k = 1, 2, ..., n;

[0244] d2 can be represented as follows:

[0245] d2=max(d 21 ,d 22 );

[0246] By converting processed text and data assets into vectors and calculating their cosine similarity, the semantic relevance between them can be measured more accurately. Furthermore, by ranking the final similarity scores of the data assets, the most relevant results are prioritized, helping users find the information they need more quickly.

[0247] This embodiment, through the above-described scheme, obtains the user's search content; recalls data assets corresponding to the search content based on a preset search strategy, obtaining several recall results. The search strategy is set in combination with several dimensions. Specifically, the search content is preprocessed to obtain processed text; the data assets corresponding to the processed text are recalled based on the preset search strategy, obtaining the several recall results. The search strategy is set in combination with the several dimensions. Specifically, it is determined whether the search content contains spaces; if the search content does not contain spaces, the search content is segmented into words to obtain the processed text; if the search content contains spaces, the search content is divided into multiple keywords based on the spaces; the keywords are segmented into words, and the keywords and the segmented keywords are used as the processed text. Specifically, the following steps are performed: matching the processed text with the names of each data asset to obtain a first recall result; and / or determining whether the processed text contains Chinese names; if the processed text contains Chinese names, performing semantic similarity retrieval on the processed text and each of the data assets to obtain a second recall result; and / or recalling the historical browsing assets of the user's department colleagues to obtain a third recall result; and / or recalling data assets related to the user's historical browsing assets to obtain a fourth recall result; and / or recalling data assets with browsing volume greater than a preset browsing volume threshold and / or authorization volume greater than a preset authorization volume threshold to obtain a fifth recall result. Further, the processed text is converted into a processed text vector; the names of each data asset are converted into name vectors; the cosine similarity between the processed text vector and the name vectors of each data asset is calculated to obtain a first similarity for each data asset; the similarity between the processed text vector and each field in each data asset is calculated; the maximum similarity between the processed text vector and each field in each data asset is taken as a second similarity for each data asset; the first similarity and the second similarity of the data assets are compared, and the larger value is taken as the final similarity of the data asset; the final similarities of each data asset are sorted to obtain a sorting result; the sorting result is selected and recalled based on a preset sorting result selection strategy to obtain a second recall result.

[0248] This invention, based on a pre-defined search strategy combining several dimensions, retrieves data assets corresponding to search content, yielding retrieved results. By optimizing the retrieved content across multiple dimensions, the results better align with user intent and meet user needs in specific usage scenarios. Specifically, preprocessing the search content reduces interference and noise, improving the accuracy of the retrieved results. Furthermore, preprocessing simplifies the search content, removing redundant information and complex structures, thus reducing search complexity and improving efficiency. By segmenting the search content into more representative words, long strings can be divided into more precise matching of relevant data assets, improving search result accuracy. For search content containing spaces, segmentation based on spaces and subsequent keyword segmentation transforms the content into more easily understood and processed text, providing more flexible search results. Finally, by considering five dimensions—text matching, text similarity, browsing by colleagues within the same department, and user's historical search content—this invention increases the coverage of retrieved results and improves the recall rate, helping users find more relevant data assets. It can also optimize the retrieved content, making search results more aligned with user intent. Furthermore, higher data asset pageview and authorization numbers indicate more searches and access to the asset, suggesting its importance; therefore, retrieving these data assets optimizes the retrieval results. Further, by converting processed text and data assets into vectors and calculating their cosine similarity, this embodiment of the invention can more accurately measure their semantic relevance. Moreover, by ranking the final similarity scores of data assets, the most relevant results are prioritized, helping users find the information they need more quickly.

[0249] Reference Figure 8 , Figure 8 This is a flowchart illustrating another exemplary embodiment of the data search method of the present invention.

[0250] As one embodiment, step S313, and / or, recalling the historical browsing assets of the user's department colleagues to obtain a third recall result includes:

[0251] Step S3131: Obtain the employee ID of the user.

[0252] Step S3132: Based on the user's employee ID, query the employee IDs of the user's department colleagues.

[0253] One implementation method is to use a user master table to store information about all employees, including employee ID information and department information.

[0254] Then, the ID of the department mentioned by the employee can be retrieved from the user master table based on the employee ID.

[0255] Then, based on the ID of the department mentioned by the employee, the IDs of all the employee's colleagues in all departments are retrieved.

[0256] Step S3133: Based on the employee ID of the colleague in the department, query the historical browsing assets of the colleague in the department, and retrieve the third recall result.

[0257] Employees' historical browsing assets can be found in their browsing history.

[0258] Therefore, you can retrieve a colleague's historical browsing assets by searching their browsing history using their ID, and then recall those assets.

[0259] In this context, due to the strong business connections among colleagues within the same department, a user's target data assets are highly likely to have also been viewed by other colleagues in the department. Therefore, recalling the historical browsing assets of department colleagues can optimize the recall results to some extent. The formula for calculating the score d3 for this part is as follows, where v is the number of visits by department colleagues in the past month:

[0260] d3 = log(v+1);

[0261] This embodiment, through the above-described scheme, obtains the user's search content; recalls data assets corresponding to the search content based on a preset search strategy, obtaining several recall results. The search strategy is set in combination with several dimensions. Specifically, the search content is preprocessed to obtain processed text; the data assets corresponding to the processed text are recalled based on the preset search strategy, obtaining the several recall results. The search strategy is set in combination with the several dimensions. Specifically, it is determined whether the search content contains spaces; if the search content does not contain spaces, the search content is segmented into words to obtain the processed text; if the search content contains spaces, the search content is divided into multiple keywords based on the spaces; the keywords are segmented into words, and the keywords and the segmented keywords are used as the processed text. Specifically, the process involves matching the processed text with the names of various data assets to obtain a first recall result; and / or determining whether the processed text contains Chinese names. If the processed text contains Chinese names, semantic similarity retrieval is performed on the processed text and each of the data assets to obtain a second recall result; and / or recalling the historical browsing assets of the user's department colleagues to obtain a third recall result; and / or recalling data assets related to the user's historical browsing assets to obtain a fourth recall result; and / or recalling data assets with browsing volume exceeding a preset browsing volume threshold and / or authorization volume exceeding a preset authorization volume threshold to obtain a fifth recall result. Specifically, the process includes obtaining the user's employee ID; querying the employee ID of the user's department colleagues based on the user's employee ID; querying the historical browsing assets of the department colleagues based on the department colleagues' employee IDs, and recalling them to obtain the third recall result.

[0262] This invention, based on a pre-defined search strategy combining several dimensions, retrieves data assets corresponding to search content, yielding retrieved results. By optimizing the retrieved content across multiple dimensions, the results better align with user intent and meet user needs in specific usage scenarios. Specifically, preprocessing the search content reduces interference and noise, improving the accuracy of the retrieved results. Furthermore, preprocessing simplifies the search content, removing redundant information and complex structures, thus reducing search complexity and improving efficiency. By segmenting the search content into more representative words, long strings can be divided into more precise matching of relevant data assets, improving search result accuracy. For search content containing spaces, segmentation based on spaces and subsequent keyword segmentation transforms the content into more easily understood and processed text, providing more flexible search results. Finally, by considering five dimensions—text matching, text similarity, browsing by colleagues within the same department, and user's historical search content—this invention increases the coverage of retrieved results and improves the recall rate, helping users find more relevant data assets. It can also optimize the retrieved content, making search results more aligned with user intent. Furthermore, higher data asset pageview and authorization numbers indicate more searches and accesses to that asset, suggesting its importance. Therefore, retrieving these data assets can optimize the retrieval results. In particular, given the strong business connections among colleagues within the same department, a user's target data asset is highly likely to have also been viewed by other colleagues. Therefore, retrieving the historical browsing assets of department colleagues can, to some extent, optimize the retrieval results.

[0263] Reference Figure 9 , Figure 9 This is a flowchart illustrating another exemplary embodiment of the data search method of the present invention.

[0264] As one embodiment, step S314, and / or recalling data assets related to the user's historical browsing assets to obtain a fourth recall result includes:

[0265] Step S3141: Obtain the employee ID of the user.

[0266] Step S3142: Based on the user's employee ID, retrieve the user's historical browsing assets.

[0267] Among these features, users can query their historical browsing assets based on their employee ID in the browsing history.

[0268] Step S3143: Based on the user's historical browsing assets, find the upstream and downstream assets related to the user's historical browsing assets.

[0269] One possible approach is to search for only one level of blood relation.

[0270] Finding only one level of kinship means searching only for directly related assets, without expanding to deeper levels of association.

[0271] Step S3144: Recall the upstream and downstream assets related to the blood relationship to obtain the fourth recall result.

[0272] The sheer number and complexity of data asset names, coupled with their strong correlation to project names and personal naming habits, makes them difficult for users to remember completely. By incorporating the relevance of assets viewed in the user's history, the burden on memory can be reduced, making it easier to find and revisit previously viewed assets. Furthermore, some data assets are crucial and used in different projects, leading to repeated searches for the same asset. Additionally, related data assets exhibit strong correlation; therefore, recalling previously viewed and related assets optimizes the retrieval results. Here, 'b' represents the number of times a data asset is related to assets viewed in the user's history.

[0273] d4 = log(b+1)

[0274] This embodiment, through the above-described scheme, obtains the user's search content; recalls data assets corresponding to the search content based on a preset search strategy, obtaining several recall results. The search strategy is set in combination with several dimensions. Specifically, the search content is preprocessed to obtain processed text; the data assets corresponding to the processed text are recalled based on the preset search strategy, obtaining the several recall results. The search strategy is set in combination with the several dimensions. Specifically, it is determined whether the search content contains spaces; if the search content does not contain spaces, the search content is segmented into words to obtain the processed text; if the search content contains spaces, the search content is divided into multiple keywords based on the spaces; the keywords are segmented into words, and the keywords and the segmented keywords are used as the processed text. The process involves: matching the processed text with the names of various data assets to obtain a first recall result; and / or determining whether the processed text contains Chinese names. If the processed text contains Chinese names, semantic similarity retrieval is performed on the processed text and each of the data assets to obtain a second recall result; and / or recalling the historical browsing assets of the user's department colleagues to obtain a third recall result; and / or recalling data assets related to the user's historical browsing assets to obtain a fourth recall result; and / or recalling data assets with browsing volume exceeding a preset browsing volume threshold and / or authorization volume exceeding a preset authorization volume threshold to obtain a fifth recall result. Specifically, the process includes: obtaining the user's employee ID; finding the user's historical browsing assets based on the user's employee ID; finding upstream and downstream assets related to the user's historical browsing assets based on the user's historical browsing assets; and recalling the upstream and downstream assets related to the historical browsing assets to obtain the fourth recall result.

[0275] This invention, based on a pre-defined search strategy combining several dimensions, retrieves data assets corresponding to search content, yielding retrieved results. By optimizing the retrieved content across multiple dimensions, the results better align with user intent and meet user needs in specific usage scenarios. Specifically, preprocessing the search content reduces interference and noise, improving the accuracy of the retrieved results. Furthermore, preprocessing simplifies the search content, removing redundant information and complex structures, thus reducing search complexity and improving efficiency. By segmenting the search content into more representative words, long strings can be divided into more precise matching of relevant data assets, improving search result accuracy. For search content containing spaces, segmentation based on spaces and subsequent keyword segmentation transforms the content into more easily understood and processed text, providing more flexible search results. Finally, by considering five dimensions—text matching, text similarity, browsing by colleagues within the same department, and user's historical search content—this invention increases the coverage of retrieved results and improves the recall rate, helping users find more relevant data assets. It can also optimize the retrieved content, making search results more aligned with user intent. Furthermore, higher data asset pageview and authorization numbers indicate more searches and access to the asset, suggesting its importance. Retrieving these data assets optimizes the retrieval results. Since data asset names are numerous and complex, and their naming is highly correlated with project names and personal naming habits, they are difficult for users to remember completely. Combining the relevance of users' historical asset browsing history reduces the burden on users' memory, making it easier for them to find and revisit previously viewed related assets. Additionally, some data assets are important and used by users in different projects, leading to repeated searches for the same asset. Furthermore, related data assets have strong connections; therefore, retrieving assets from users' historical browsing history and those with related connections optimizes the retrieval results.

[0276] Reference Figure 10 , Figure 10 This is a schematic diagram illustrating the overall process of data search in an embodiment of the present invention.

[0277] Step S10001: User inputs content;

[0278] Step S10002: Preprocess the user input to obtain the processed text;

[0279] Step S10003: Use the first search strategy to search for data assets based on the processed text, specifically by matching the processed text with the data assets (direct matching, indirect matching) to obtain the first recall result.

[0280] Step S10004: Use the second search strategy to search for data assets based on the processed text, specifically by comparing the similarity between the processed text and the data assets (asset name similarity, field similarity) to obtain the second recall result.

[0281] Step S10005: Use the third search strategy to search for data assets. Specifically, recall data assets whose historical search volume is greater than a preset threshold, and recall data assets whose department colleagues' browsing volume is greater than a preset threshold, to obtain the third recall result.

[0282] Step S10006: Use the fourth search strategy to search for data assets, specifically to recall the first-level parent assets of the user's historical browsing assets and the first-level child assets of the user's historical browsing assets, and obtain the fourth recall result.

[0283] Step S10007: Use the fifth search strategy to search for data assets, specifically to recall data assets with more than a preset threshold for page views and more than a preset threshold for authorization, and obtain the fifth recall result.

[0284] Step S10008: Input the above-mentioned recall results into the weighting layer and assign different weights to different recall results;

[0285] Step S10009: In the output layer, set the display order of the recall results according to different weights.

[0286] This embodiment, through the above-described scheme, obtains the user's search content; based on a preset search strategy, it retrieves the data assets corresponding to the search content, obtaining several retrieved results. The search strategy is obtained by combining several dimensions. This embodiment of the invention retrieves the data assets corresponding to the search content based on a preset search strategy combined with several dimensions, obtaining retrieved results. By optimizing the retrieved content by combining multiple dimensions, the retrieved results can better match the user's intent and meet the user's needs in specific usage scenarios.

[0287] Furthermore, this application also proposes an apparatus comprising a memory, a processor, and a data search program stored in the memory and executable on the processor, wherein the data search program, when executed by the processor, implements the steps of the data search method described above.

[0288] Since this data search program employs all the technical solutions of all the foregoing embodiments when executed by the processor, it has at least all the beneficial effects brought about by all the technical solutions of all the foregoing embodiments, which will not be elaborated here.

[0289] Furthermore, embodiments of this application also propose a computer-readable storage medium storing a data search program, which, when executed by a processor, implements the steps of the data search method described above.

[0290] Since this data search program employs all the technical solutions of all the foregoing embodiments when executed by the processor, it has at least all the beneficial effects brought about by all the technical solutions of all the foregoing embodiments, which will not be elaborated here.

[0291] This embodiment, through the above-described scheme, specifically obtains the user's search content; based on a preset search strategy, it retrieves data assets corresponding to the search content, obtaining several retrieved results. The search strategy is obtained by combining several dimensions. This embodiment of the invention retrieves data assets corresponding to the search content based on a preset search strategy combined with several dimensions, obtaining retrieved results. By optimizing the retrieved content by combining multiple dimensions, the retrieved results can be made more in line with user intent and meet user needs in specific usage scenarios.

[0292] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or approach that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or approach. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or approach that includes that element.

[0293] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0294] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, controlled terminal, or network device, etc.) to execute the methods of each embodiment of the present invention.

[0295] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A data search method, characterized in that, The method includes the following steps: Obtain the user's search content; Based on a preset search strategy, data assets corresponding to the search content are retrieved to obtain several retrieval results. The search strategy is obtained by combining several dimensions. The step of obtaining the user's search content is followed by: The search content is preprocessed to obtain processed text, specifically including: Determine whether the search content contains spaces; If the search content does not contain spaces, then the search content is segmented into words to obtain the processed text; If the search content contains the space, then the search content is divided into multiple keywords based on the space; The keywords are segmented into words, and the keywords and the segmented keywords are used as the processed text. The step of recalling data assets corresponding to the search content based on a preset search strategy to obtain several recall results includes: Based on the preset search strategy, data assets corresponding to the processed text are retrieved to obtain the plurality of retrieval results. The search strategy is obtained by combining the plurality of dimensions and specifically includes: The processed text is matched with the names of each data asset to obtain the first recall result; Determine whether the processed text contains Chinese names. If the processed text contains Chinese names, perform semantic similarity retrieval on the processed text and each of the data assets to obtain a second recall result. The historical browsing assets of the user's department colleagues are recalled to obtain a third recall result, wherein the historical browsing assets of the department colleagues are the data assets that the department colleagues have previously browsed; Data assets that are related to the user's historical browsing assets are recalled to obtain the fourth recall result. Among them, data assets that are related to the user's historical browsing assets are data assets that are associated with or connected to the user's historical browsing assets. Data assets with pageviews exceeding a preset pageview threshold and / or authorizations exceeding a preset authorization threshold are recalled, resulting in the fifth recall result; The step of recalling the historical browsing assets of the user's department colleagues to obtain a third recall result includes: Obtain the user's employee ID; Based on the user's employee ID, the employee IDs of the user's department colleagues can be retrieved; Based on the employee ID of the colleague in the department, the historical browsing assets of the colleague in the department are obtained, and the third recall result is obtained; The step of recalling data assets related to the user's historical browsing assets to obtain the fourth recall result includes: Obtain the user's employee ID; The user's historical browsing assets can be retrieved based on the user's employee ID; Based on the user's historical browsing assets, find the upstream and downstream assets related to the user's historical browsing assets; The upstream and downstream assets related to the blood relationship are recalled, resulting in the fourth recall result.

2. The method according to claim 1, characterized in that, The step of recalling data assets corresponding to the search content based on a preset search strategy and obtaining several recall results further includes: The data asset score values ​​of the several recall results are calculated based on a preset weight allocation strategy; Based on the data asset scores of the recall results, the recall results are sorted, and the sorted recall results are returned to the user.

3. The method according to claim 1, characterized in that, The step of performing semantic similarity retrieval on the processed text and each of the data assets to obtain the second recall result includes: Convert the processed text into a processed text vector; Convert the names of each data asset into name vectors for each data asset. Calculate the cosine similarity between the processed text vector and the name vector of each data asset to obtain the first similarity of each data asset; The similarity between the processed text vector and each field in each of the data assets is calculated. The maximum similarity between the processed text vector and each field in each of the data assets is taken as the second similarity of each of the data assets; The first similarity and the second similarity of the data assets are compared, and the larger value is taken as the final similarity of the data assets. The final similarity scores of each of the data assets are sorted to obtain a sorting result; The sorting results are selected and recalled based on a preset sorting result selection strategy to obtain the second recall result.

4. A data search device, characterized in that, The data search device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the data search method as described in any one of claims 1-3.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the data search method as described in any one of claims 1-3.

Images