A method for evaluating the protection effect of historical and cultural blocks based on big data

Abstract

This specification provides a method for evaluating the effectiveness of historical and cultural block preservation based on big data. This invention relates to the fields of smart city and digital preservation of historical and cultural heritage. It generates a target element pool for each block based on a block name list, task prompts, and various knowledge bases, thus expanding the information in the block name list and ensuring the completeness of the block data collected from target multimodal data. From the various dimensions of the target multimodal data, dimensions for evaluating the effectiveness of block preservation are determined as original evaluation indicators to comprehensively cover the different dimensions of data for each block. Based on these original evaluation indicators, principal component analysis is used to determine the target evaluation indicators, constructing a multi-level indicator system for evaluating the effectiveness of block preservation. The preservation effectiveness of each block is then evaluated, improving the accuracy of the evaluation results.

Description

Technical Field

[0001] This specification relates to the fields of smart city and digital protection technology of historical and cultural heritage, and in particular to an evaluation method for the effectiveness of historical and cultural block protection based on big data. Background Technology

[0002] Utilizing digital and intelligent technologies for refined management of historical and cultural blocks is a core approach to balancing protection, inheritance, and revitalization. The implementation of dynamic monitoring and early warning mechanisms can overcome the limitations of traditional management's lag and subjectivity, providing scientific support for the long-term protection of historical and cultural blocks. However, existing methods for evaluating the effectiveness of historical and cultural block protection have the following limitations:

[0003] First, existing assessment methods primarily collect monomodal data related to historical and cultural blocks and then evaluate these blocks using pre-constructed indicator systems. For example, they utilize location big data, points of interest (POIs), and internet platform comment data as monomodal data to analyze the usage characteristics, business composition, and popularity of historical and cultural blocks. However, the data used in these methods is monomodal, which significantly underutilizes the rich perceptual information from new media platforms, such as videos and images, making it difficult to comprehensively capture the public's authentic and multi-dimensional feedback on the blocks. Furthermore, analyzing only monomodal data results in a limited analytical dimension, failing to integrate multimodal data analysis and thus failing to fully reflect the social recognition of the conservation effectiveness. Second, existing assessment methods mainly collect data based on the block names corresponding to historical and cultural blocks, easily overlooking other name information associated with these blocks and failing to comprehensively cover all related entities within the historical and cultural blocks (such as historical buildings), leading to insufficient data completeness. Third, due to the cost and technical barriers of data collection and analysis, existing assessment methods mainly analyze and assess individual historical and cultural blocks, lacking a quantifiable assessment system that can be used to assess the protection effectiveness of multiple historical and cultural blocks. This makes it difficult to effectively compare assessment results among blocks of different types and regions, limiting its universality and policy guidance value.

[0004] Based on this, this specification provides a method for evaluating the effectiveness of historical and cultural block protection based on big data. Summary of the Invention

[0005] To address the shortcomings of existing assessment methods, such as limited data modalities, single analytical dimensions, insufficient data completeness, and narrow assessment scope, this specification provides a big data-based method for evaluating the effectiveness of historical and cultural district protection. It enhances the generation of a large-scale model of district elements through multi-source knowledge data retrieval, generating a target element pool encompassing various categories of district elements. This expands the information on the list of district names, ensuring the completeness of subsequently collected district data. Using principal component analysis based on the target multimodal data, a multi-level indicator system for evaluating the effectiveness of district protection is generated, enabling a more comprehensive assessment of the protection outcomes of historical and cultural districts.

[0006] The following technical solution is adopted in this specification:

[0007] This specification provides a method for evaluating the effectiveness of historical and cultural district protection based on big data. The method includes:

[0008] S1: Generate street elements based on a large model: Obtain a list of street names and task prompts and input them into a large model for generating street elements. Enhance the large model for generating street elements by searching through pre-built knowledge bases. Generate street elements for each street object in the list of street names through the large model for generating street elements. Based on the street name, entity elements, spatial elements and social element categories, combine them into a target element pool as the output of the large model for generating street elements.

[0009] S2: Obtain street data from multimodal data: Obtain target multimodal data from the data source based on different data processing strategies; Collect data matching the street elements in the target element pool of each street object from the target multimodal data, and use it as street data;

[0010] S3: Construct a multi-level indicator system for evaluating the effectiveness of neighborhood protection: From the various dimensions of the target multimodal data, determine the dimensions used to evaluate the effectiveness of neighborhood protection as each original evaluation indicator, and based on the original evaluation indicators, use principal component analysis to determine each target evaluation indicator, and construct a multi-level indicator system for evaluating the effectiveness of neighborhood protection.

[0011] S4: Evaluate the effectiveness of the protection of historical and cultural blocks: Based on the block data and the multi-level indicator system for evaluating the effectiveness of block protection, evaluate the effectiveness of the protection of each block.

[0012] Optionally, the knowledge base includes a geographic information database, a neighborhood encyclopedia, a historical document database, and a conservation planning knowledge base.

[0013] Optionally, S1 specifically includes:

[0014] Obtain a list of neighborhood names and task prompts;

[0015] The block name list and the task prompt words are input into the block element generation model. The block element generation model is enhanced by searching through pre-built knowledge bases. The block element generation model generates block elements for each block object in the block name list. The block elements of each block object are divided based on block name, entity elements, spatial elements and social element categories. The results of the division of each block object are combined into the initial element pool of each block object and used as the output of the block element generation model.

[0016] Perform feature cleaning on the block features in the initial feature pool of each block object;

[0017] The knowledge verification model is used to verify the street elements in each initial element pool after cleaning, so as to obtain the target element pool for each street object.

[0018] Optionally, the multi-level indicator system for evaluating the effectiveness of street protection includes each primary evaluation indicator, each secondary evaluation indicator, and each tertiary evaluation indicator. Each primary evaluation indicator includes a first number of secondary evaluation indicators, and each secondary evaluation indicator includes a second number of tertiary evaluation indicators.

[0019] Optionally, S4 specifically includes:

[0020] S41: Based on the street data of each street object, determine the index value of each third-level evaluation index corresponding to each street object;

[0021] S42: Construct the original matrix of the three-level evaluation indicators based on the indicator values ​​of each of the three-level evaluation indicators for each block object;

[0022] S43: Based on the original matrix of the three-level evaluation indicators, determine the target indicator weight corresponding to each evaluation indicator in the multi-level indicator system for evaluating the effectiveness of street protection;

[0023] S44: Determine the protection effectiveness index corresponding to each block object based on the target indicator weights corresponding to each evaluation indicator and the original matrix of the three-level evaluation indicators.

[0024] Optionally, S43 specifically includes:

[0025] S431: The three-level evaluation indicators are used as evaluation indicators to be processed, and the original matrix of the three-level evaluation indicators is used as the original matrix of evaluation indicators to be processed.

[0026] S432: Regularize the original matrix of the evaluation indicators to be processed to obtain the standardized matrix of the evaluation indicators to be processed.

[0027] S433: Based on the standardized matrix of the evaluation indicators to be processed, the entropy value corresponding to each evaluation indicator to be processed is calculated using the entropy method.

[0028] S434: Determine the difference coefficient corresponding to each evaluation indicator to be processed based on the entropy value corresponding to each evaluation indicator to be processed;

[0029] S435: Normalize the difference coefficients of each evaluation indicator to be processed under the same superior evaluation indicator to obtain the target indicator weights of each evaluation indicator to be processed under the same superior evaluation indicator.

[0030] S436: When the evaluation indicator to be processed is not the first-level evaluation indicator, determine the indicator value corresponding to the superior evaluation indicator of each evaluation indicator to be processed for each block object according to the target indicator weight corresponding to each evaluation indicator to be processed and the standardized matrix of the evaluation indicator to be processed, and construct the original matrix of the superior evaluation indicator.

[0031] S437: The superior evaluation indicator is used as the evaluation indicator to be processed again, and the original matrix of the superior evaluation indicator is used as the original matrix of the evaluation indicator to be processed again. Steps S432 to S437 are repeated until the evaluation indicator to be processed is a first-level evaluation indicator. The target indicator weights corresponding to each evaluation indicator in the multi-level indicator system for evaluating the effectiveness of the protection of the street are obtained.

[0032] Optionally, S432 specifically includes:

[0033] When the original matrix of the evaluation indicators to be processed is the original matrix of the third-level evaluation indicators, a preset first regularization formula is used to regularize the indicator values ​​corresponding to the positive evaluation indicators in the original matrix of the evaluation indicators to be processed, and a preset second regularization formula is used to regularize the indicator values ​​corresponding to the negative evaluation indicators in the original matrix of the evaluation indicators to be processed, so as to obtain the standardized matrix of the evaluation indicators to be processed.

[0034] When the original matrix of the evaluation indicators to be processed is not the original matrix of the third-level evaluation indicators, the first regularization formula is used to regularize the indicator values ​​corresponding to all evaluation indicators in the original matrix of the evaluation indicators to be processed.

[0035] Optionally, S435 specifically includes:

[0036] When the evaluation index to be processed is the secondary evaluation index, the difference coefficients corresponding to each evaluation index to be processed under the same superior evaluation index are normalized to determine the initial index weights corresponding to each evaluation index to be processed under the same superior evaluation index.

[0037] Based on the Delphi method, the initial indicator weights corresponding to the first evaluation indicator among the evaluation indicators to be processed are adjusted to obtain the target indicator weights corresponding to the first evaluation indicator; the first evaluation indicator is determined according to the importance level and data fluctuation of each evaluation indicator to be processed.

[0038] Identify the assessment indicators to be processed that belong to the same superior assessment indicator as the first assessment indicator, and use them as related assessment indicators.

[0039] Based on the target indicator weight corresponding to the first evaluation indicator, the initial indicator weight corresponding to the associated evaluation indicator is adjusted to obtain the target indicator weight corresponding to the associated evaluation indicator.

[0040] The initial indicator weights corresponding to the other evaluation indicators to be processed, excluding the first evaluation indicator and the associated evaluation indicator, are used as the target indicator weights.

[0041] Optionally, determining the protection effectiveness index for each street block object in step S44, based on the target indicator weights corresponding to each evaluation indicator and the original matrix of the three-level evaluation indicators, specifically includes:

[0042] The original matrix of the three-level evaluation indicators is regularized to obtain the standardized matrix of the three-level evaluation indicators.

[0043] Based on the standardized matrix of the three-level evaluation indicators, the third indicator value corresponding to each of the three-level evaluation indicators for each block object is determined;

[0044] Using the target indicator weight of each of the three-level evaluation indicators, the third indicator value of each of the three-level evaluation indicators of each block object is weighted to obtain the third weighted value of each of the three-level evaluation indicators of each block object;

[0045] Based on the third weighted value of each of the three-level evaluation indicators for each block object, determine the second indicator value of each of the two-level evaluation indicators for each block object;

[0046] Using the target indicator weight of each secondary evaluation indicator, the second indicator value of each secondary evaluation indicator of each block object is weighted to obtain the second weighted value of each secondary evaluation indicator of each block object;

[0047] Based on the second weighted value of each secondary evaluation indicator for each block object, determine the first indicator value of each primary evaluation indicator for each block object;

[0048] Using the target indicator weight of each primary evaluation indicator, the first indicator value of each primary evaluation indicator of each block object is weighted to obtain the first weighted value of each primary evaluation indicator of each block object;

[0049] The sum of the first weighted values ​​of each primary evaluation indicator for each block object is determined and used as the corresponding protection effectiveness index for each block object.

[0050] Optionally, the step S2, which involves obtaining target multimodal data from the data source based on different data processing strategies, specifically includes:

[0051] We acquire data of various modalities from various types of data sources and use it as raw multimodal data.

[0052] Based on different data processing strategies, the multimodal data is processed to obtain target multimodal data; the data processing strategies include data cleaning, data semantic processing, media rating classification, sentiment analysis, and event extraction and recognition.

[0053] Optionally, in step S3, based on the original evaluation indicators, principal component analysis is used to determine the target evaluation indicators and construct a multi-level indicator system for evaluating the effectiveness of street protection. This specifically includes:

[0054] Based on the target multimodal data, determine the original indicator value corresponding to each original evaluation indicator of each block object;

[0055] Determine the original data matrix consisting of the original index values ​​corresponding to each original evaluation index of each block object;

[0056] Each data element in the original data matrix is ​​standardized to obtain the target data matrix;

[0057] Based on the target data matrix, the correlation coefficient matrix is ​​calculated using principal component analysis, and the correlation coefficient matrix is ​​decomposed to obtain the eigenvalues ​​and eigenvectors corresponding to each principal component.

[0058] Based on the eigenvalues ​​of each principal component, the principal components are screened to obtain the target principal components;

[0059] Identify the elements in the eigenvector corresponding to each target principal component whose loading reaches a specified threshold, and use them as target elements;

[0060] Determine the original evaluation indicators corresponding to each target element, and use them as the evaluation indicators for each target;

[0061] Based on the aforementioned target evaluation indicators, a multi-level indicator system for evaluating the effectiveness of neighborhood protection is constructed.

[0062] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects:

[0063] This manual provides a method for evaluating the effectiveness of historical and cultural district protection based on big data. It begins by obtaining a list of district names and task prompts, which are then input into a large-scale district element generation model. This model is enhanced through retrieval from various knowledge bases. The model generates district elements for each district in the district name list, categorizing them based on district name, entity elements, spatial elements, and social elements. These elements are combined into a target element pool, which serves as the output of the large-scale district element generation model. Further enhancement through multi-source knowledge data retrieval generates a target element pool for each district, encompassing multiple categories of district elements. This expands the information on the district name list, preventing the omission of other names corresponding to historical and cultural districts and comprehensively covering all related entities within the historical and cultural district (such as historical buildings), thus ensuring the completeness of subsequent district data collected based on the target element pool. Then, using different data processing strategies, target multimodal data is obtained from the data source. This helps to comprehensively capture the public's real and multi-dimensional feedback on the districts, fully reflecting the social recognition of the protection effectiveness and providing data support for subsequent evaluation of protection effectiveness. From the target multimodal data, data matching the district elements in the target element pool for each district is collected and used as district data. Then, from the various dimensions of the target multimodal data, each dimension used to evaluate the effectiveness of street block protection is determined as a primary evaluation indicator. Based on these primary indicators, principal component analysis is used to determine the target evaluation indicators and construct a multi-level indicator system for evaluating the effectiveness of street block protection. By constructing primary evaluation indicators based on the data dimensions included in the target multimodal data, the primary evaluation indicators comprehensively cover the different dimensions of data for each street block. Principal component analysis is then used to filter the primary evaluation indicators to select those that more concisely and essentially reflect the effectiveness of street block protection while preserving the information from the target multimodal data to the greatest extent possible. Based on the street block data and the multi-level indicator system for evaluating the effectiveness of street block protection, the protection effectiveness of each street block is evaluated to improve the accuracy of the evaluation results and provide a scientific tool for the refined governance of historical and cultural streets.

[0064] In order to ensure the accuracy of the generated street elements, the street elements in the initial element pool generated by the large model for generating street elements can be cleaned, and the street elements in each initial element pool after cleaning can be verified by the knowledge verification model, so as to improve the accuracy of the street data collected subsequently based on the target element pool.

[0065] The multi-level indicator system for evaluating the effectiveness of street block protection in this invention may include primary evaluation indicators, secondary evaluation indicators, and tertiary evaluation indicators. Each primary evaluation indicator includes a first number of secondary evaluation indicators, and each secondary evaluation indicator includes a second number of tertiary evaluation indicators. When evaluating the protection effectiveness of each street block, the entropy method can be used first to calculate the target indicator weight for each evaluation indicator. Then, using the target indicator weights and street block data, the protection effectiveness index corresponding to each street block can be determined to complete the evaluation of the protection effectiveness of each street block, achieving objective quantification and horizontal comparative evaluation of the protection effectiveness of different historical and cultural street blocks.

[0066] Furthermore, since there may be situations where assessment indicators with high importance and low data fluctuation have low indicator weights, subjective weights can be assigned to assessment indicators with high importance and low data fluctuation (the first assessment indicator) using the Delphi method. Based on these subjective weights, the indicator weights of the assessment indicators to be processed that belong to the same superior assessment indicator as the first assessment indicator can be adjusted to avoid situations where assessment indicators with high importance and low data fluctuation have low indicator weights, and also to avoid affecting the accuracy of the protection effectiveness assessment results.

[0067] In order to facilitate the collection of street data, the present invention can first perform data cleaning, semantic processing, media level classification, sentiment analysis and event extraction and recognition on the modal data (i.e., raw multimodal data) obtained from various types of data sources according to different preset data processing strategies to obtain target multimodal data. Then, street data is collected based on the target multimodal data, which helps to improve the collection speed of street data and avoids data redundancy and interference from useless data. Attached Figure Description

[0068] The accompanying drawings, which are included to provide a further understanding of this specification and form part of this specification, illustrate exemplary embodiments and are used to explain this specification, but do not constitute an undue limitation thereof. In the drawings:

[0069] Figure 1 This is a flowchart illustrating a method for evaluating the effectiveness of historical and cultural district protection based on big data, as provided in this specification.

[0070] Figure 2 This is a schematic diagram of a multi-level indicator system for evaluating the effectiveness of neighborhood protection, as provided in this specification.

[0071] Figure 3 This is a schematic diagram illustrating a process for generating street elements as provided in this specification. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.

[0073] This specification provides a method for evaluating the effectiveness of historical and cultural block protection based on big data. The technical solutions provided by each embodiment of this specification are described in detail below with reference to the accompanying drawings.

[0074] Figure 1 This document presents a flowchart illustrating a big data-based method for evaluating the effectiveness of historical and cultural district preservation, which includes the following steps:

[0075] S1: Generate street elements based on a large model: Obtain a list of street names and task prompts, and input them into a large model for generating street elements. Enhance the large model by searching through pre-built knowledge bases. Generate street elements for each street object in the list of street names through the large model. Based on the street name, entity elements, spatial elements, and social element categories, combine them into a target element pool as the output of the large model for generating street elements.

[0076] In this manual, the equipment used for assessing the effectiveness of conservation efforts can first generate street block elements based on a large model. This involves obtaining a list of street block names and task prompts, inputting the street block elements, and generating a large model. The large model is then enhanced through pre-built knowledge bases. Finally, it generates street block elements for each street block in the list of street block names. These elements are then categorized based on street block name, entity elements, spatial elements, and social elements, and combined into a target element pool as the output of the large model. The equipment used for assessing the effectiveness of conservation efforts can be a server, a system, one or more modules within a system, or electronic devices such as desktop computers or laptops. For ease of description, the following explanation focuses on a server as the primary execution entity, illustrating a big data-based method for assessing the effectiveness of historical and cultural street block conservation.

[0077] The aforementioned list of street names is pre-built and includes basic information on multiple historical and cultural streets nationwide. The specific number of historical and cultural streets may be updated in real time based on actual circumstances. This instruction manual uses 1241 historical and cultural streets as an example. Furthermore, the basic information may include location information and street name information. Location information may include the province (or autonomous region) and city / county name where the historical and cultural street is located. Street name information may include the original street name and the adjusted street name. The original street name is the historical or folk name of the historical and cultural street, and the adjusted street name is the official name of the historical and cultural street. It should be noted that the above street name information may be updated in real time according to actual circumstances. The street objects mentioned above refer to historical and cultural streets.

[0078] The task prompts mentioned above are pre-set, but can be manually constructed. These prompts guide the large-scale street element generation model to generate an element pool containing multiple categories of street elements for each street object in the street name list, based on the input and knowledge data from various knowledge bases. The large-scale street element generation model is pre-set with hundreds of billions of parameters. This model can be the Tuotian large-scale model, or any other large-scale model capable of generating street elements, or a finely tuned version of a general large-scale model; this specification does not impose specific limitations. The Tuotian large-scale model is based on a general large-scale model, pre-trained using hundreds of billions of internet media information data from a pre-built internet data center, and undergoes reinforcement learning and model distillation. It also incorporates various technical components such as RAG (Retrieval-Augmented Generation) and can access various types of knowledge bases. Taking the large-scale model for generating street elements as an example, which is a fine-tuned version of the general large-scale model, during fine-tuning training, fine-tuning samples and fine-tuning labels can be obtained first. The fine-tuning samples can be the basic information corresponding to historical and cultural blocks, and the fine-tuning labels are the labeled element pools of various categories of street elements corresponding to historical and cultural blocks in the fine-tuning samples, obtained through pre-manual annotation. The fine-tuning samples and task prompts are used as input to the general large-scale model. The general large-scale model is then enhanced by searching pre-built knowledge bases to generate a predicted element pool for each historical and cultural block in the fine-tuning samples. This predicted element pool includes various categories of street elements. Based on the predicted element pools and fine-tuning labels, a pre-set element loss function is used to determine the element loss. Based on the element loss, the general large-scale model is fine-tuned to obtain the large-scale model for generating street elements. The aforementioned general large-scale model can be any existing large-scale model with hundreds of billions of parameters. The aforementioned element loss function is pre-set and can be the cross-entropy loss function or other supervised loss functions; this description does not specify any particular limitation.

[0079] The aforementioned knowledge base is pre-built and may include a geographic information database, a neighborhood encyclopedia, a historical document database, and a conservation planning knowledge base. The geographic information database may include the location (e.g., latitude and longitude) of each historical and cultural neighborhood, and the location of entities within each historical and cultural neighborhood (e.g., historical buildings, traditional residences, public buildings, modern buildings, etc.). The neighborhood encyclopedia may include multi-dimensional information on the history, architecture, culture, space, and current conservation status of each historical and cultural neighborhood. Specifically, it may include basic information, historical evolution, space and architecture, and culture and society. Basic information may include the neighborhood name and geographical location (latitude and longitude, administrative division). The protection level includes national, provincial, and municipal levels; historical evolution includes the founding date and major historical events; spatial and architectural aspects include street layout and building types; cultural and social aspects include intangible cultural heritage projects (festivals, folk customs, crafts, etc.) and traditional industries (time-honored brands, handicrafts, etc.); historical document database includes document information corresponding to each historical and cultural block, specifically including document title, publication date, author, and content; and protection planning knowledge base includes protection plans corresponding to each historical and cultural block, which may include information such as current status assessment, protection scope delineation, protection objects, protection types, and protection measures.

[0080] The aforementioned knowledge bases enhance the retrieval of the large-scale street element generation model by retrieving knowledge data from various knowledge bases during the generation of street elements. Based on this knowledge data, a list of street names, and task prompts, an element pool encompassing various categories of street elements is generated, allowing the large-scale street element generation model to perform retrieval as it generates elements. Furthermore, the large-scale street element generation model first generates the street elements corresponding to each street, then categorizes them based on pre-set element categories—namely, street names, entity elements, spatial elements, and social element categories—resulting in various categories of street elements. These categorized results are then combined into an element pool, the target element pool. Each street object has a corresponding categorized result, and the combination of these results for each street object forms its element pool.

[0081] The aforementioned element pools (i.e., the initial element pool and the target element pool) include various categories of street elements, including street names, physical elements, spatial elements, and social elements. Street names encompass all names corresponding to historical and cultural streets, namely historical names, folk names, and official names. Physical elements refer to the physical, perceptible, and measurable historical remains or built-up environment components within a historical and cultural street, including buildings (historical buildings, workshops, etc.), structures (ancient wells, archways, stone bridges, etc.), street paving (bluestone paths, cobblestone roads, drainage ditches, etc.), environmental features (ancient trees, water systems, rocks, etc.), and decorations (wood carvings, paintings, etc.). Spatial elements refer to the spatial structure, texture, and visual order constituted by physical elements, including geographical location, overall layout, street network, plot division, interface relationships, visual corridors, and nodes. Social elements refer to the people, communities, lifestyles, cultural practices, and collective memories that coexist with the historical and cultural street, including population structure, community organization, traditional industries, intangible cultural heritage, daily activities, and collective memories. Therefore, the above task prompts can also be used to prompt the large-scale model for generating street elements to use its own reasoning ability based on the list of street names, combined with retrieval enhancement generation technology to integrate multi-source knowledge from various knowledge bases, extract semantic elements (i.e., street elements) of streets from multiple categories such as streets, entities, spaces, and society, and combine them into an element pool as the output of the large-scale model for generating street elements.

[0082] Traditional methods for collecting data on historical and cultural blocks mainly rely on precise searches or fuzzy matching of block names. However, this approach is prone to missing information related to informal names, historical names, or colloquial names, and it is also difficult to comprehensively cover related entity elements such as historical buildings and streets within the block, resulting in insufficient data integrity. Therefore, the server can leverage the thought chain pushing capability of generating large models based on block elements, combined with retrieval-enhanced generation technology, to integrate multi-source knowledge data such as geographic information, block encyclopedias, historical document databases, and conservation planning knowledge bases. This allows for the initial extraction of block elements from multiple dimensions of "block-entity-space-society," further expanding the block names and ensuring the integrity of the data collected based on the expanded information. Specifically, the server can obtain a list of street names and task prompts, inputting these into a large-scale street element generation model. The model is then enhanced using pre-built knowledge bases; that is, the model employs retrieval enhancement techniques to retrieve knowledge data from these bases and input it into the model. This generates street elements for each street object in the street name list. The model further categorizes these elements based on street name, entity elements, spatial elements, and social element categories, resulting in a segmented pool for each street object. This pool is then combined with the target element pool for each street object and used as the output of the large-scale street element generation model. The input data for this model can include the street name list, task prompts, and various knowledge bases (or knowledge data retrieved from these bases), while the output is the element pool for each street object in the street name list. The results of the above division can be categorized into various groups, each representing a corresponding element category, namely, street name, entity element, spatial element, and social element. Each group includes street elements belonging to that element category. The above combination refers to combining the various group categories in the resulting division, that is, combining the various group categories in the division results for each street object to obtain the target element pool corresponding to each street object. The above output is the target element pool corresponding to each street object. It should be noted that retrieving knowledge data from various knowledge bases, generating street elements, dividing street elements into multiple categories, and combining the results of the division are all achieved through a large-scale street element generation model.

[0083] S2: Obtain street data from multimodal data: Obtain target multimodal data from the data source based on different data processing strategies; collect data from the target multimodal data that matches the street elements in the target element pool of each street object, and use it as street data.

[0084] In this specification, the server can acquire street data from multimodal data, that is, obtain target multimodal data from data sources based on different data processing strategies, and collect data matching street elements in the target element pool of each street object from the target multimodal data, and use this as street data. The aforementioned data sources can be of various types, including but not limited to user-generated content platforms, short video platforms, government information disclosure platforms, official media websites, or legally open government APIs. Furthermore, all of the aforementioned data sources are public, legal, and non-sensitive. Moreover, when acquiring multimodal data, various modal data can be obtained through public and legal API interfaces provided by the aforementioned various types of data sources. The data for each modality is obtained legally and compliantly or with the consent of the user and the data source provider. The aforementioned modalities include text, video, audio, and images. Specifically, when acquiring target multimodal data from data sources based on different data processing strategies, various modal data can be obtained from various types of data sources and used as raw multimodal data. Subsequently, to facilitate the retrieval of neighborhood data and the determination of corresponding indicator values ​​for evaluation metrics, the server can process the multimodal data based on different data processing strategies to obtain the target multimodal data. These data processing strategies include data cleaning, semantic data processing, media rating classification, sentiment analysis, and event extraction and recognition.

[0085] When collecting data matching the street elements in the target element pool of each street object from the target multimodal data and using it as street data, the server can use the street elements in the target element pool of each street object as search elements, retrieve data matching the search elements of that street object from the target multimodal data, and use it as street data. Specifically, data retrieval can be achieved through pre-set algorithms or models.

[0086] The data cleaning described above involves cleaning the original multimodal data based on pre-set cleaning rules. These rules may include dateline recognition, caption extraction, bottom non-text identification, and spam filtering. Specifically, the server can use dateline recognition to identify and remove information such as the issuing organization, location, and time at the beginning of news articles in the original multimodal data. The server can also use caption extraction to extract and remove captions corresponding to images in the original multimodal data; these captions are usually located above or below the images. The server can also identify and remove bottom non-text (i.e., non-core content at the end of the article) in the original multimodal data. The server can also filter spam from the original multimodal data. Through this data cleaning process, not only are large amounts of duplicate content, spam text, and garbled characters removed from the original multimodal data, but the format of the text corpus in the original multimodal data is also ensured to be uniform and standardized.

[0087] The aforementioned semantic data processing extracts text information from images using OCR recognition technology and maps different modalities of data, such as text and images, from the original multimodal data to a unified vector space (low-dimensional dense vector space) using a dual-tower model for image-text matching. This effectively captures the potential semantic relationships between different modalities, achieving semantic alignment and deep fusion of multimodal data. Specifically, the server can extract keyframes from the video data in the original multimodal data and extract text information from the extracted images and the images in the original multimodal data using OCR recognition technology. Keyframes can be pre-set. Simultaneously, the server can determine the text in the original multimodal data as each first text and the images (images and each frame in the video) as each first image. Each first text is combined with each first image to obtain image-text pairs, where each pair differs either in text or image. For each image-text pair, the text is input into the text encoder in the dual-tower model to obtain text features, and the image is input into the image encoder in the dual-tower model to obtain image features. The text and image features mentioned above are all features in a unified vector space. The matching result of the text-image pair is determined based on the distance between the text and image features. Specifically, if the distance is less than a preset value, the text-image pair is considered a match; if the distance is not less than the preset value, the text-image pair is considered a mismatch. The server can segment the text and images in the original multimodal data based on the matching result of each text-image pair, that is, group the text and images in matching text-image pairs together. The aforementioned dual-tower model can be any existing model, or a model fine-tuned from an existing model; this specification does not impose specific limitations.

[0088] The media rating classification described above categorizes the raw multimodal data into media ratings. A specific server can automatically classify the data in the raw multimodal data according to pre-set classification rules, using pre-defined algorithms or models, thus determining the media rating corresponding to each data point in the raw multimodal data. These classification rules define core media, primary media, and secondary media. The media rating is actually determined by the media of the data's data source; that is, the data corresponds to the same media rating as the data source it originates from.

[0089] The aforementioned sentiment analysis involves performing sentiment recognition on raw multimodal data to obtain sentiment recognition results, which can be positive, neutral, or negative. Specifically, for each text in the raw multimodal data, the server can use that text as input to a pre-trained sentiment classifier to determine the sentiment recognition result output by the classifier. For each image (each frame in an image or video) in the raw multimodal data, the server can determine the corresponding text information (i.e., obtained through OCR (Optical Character Recognition) technology) and use this text information as input to the sentiment classifier to determine the sentiment recognition result output by the classifier. Of course, if no corresponding text information can be recognized for the image, the sentiment recognition result corresponding to the text matching the image can be directly used as the sentiment recognition result for the image. The text matching the image can be determined through the matching results of image-text pairs, i.e., the matching result is the text in the matched image-text pair. The sentiment classifier described above is a pre-trained model, which can be a BERT (Bidirectional Encoder Representations from Transformers) model. During training, a first text sample and sentiment annotations are obtained. The sentiment annotations can include one of positive, neutral, or negative. The first text sample is input into the sentiment classifier to determine the recognition result corresponding to the first text sample. Based on the sentiment annotations and recognition results, a pre-set sentiment recognition loss function is used to determine the first loss. The sentiment classifier is then trained based on the first loss. The aforementioned sentiment recognition loss function is pre-set and can be a cross-entropy loss function or other supervised loss functions; this description does not specify any particular limitation.

[0090] The aforementioned event extraction and identification involves extracting and identifying events from the raw multimodal data to determine the negative events included within it. These negative events refer to inadequate protection of historical and cultural districts, destructive behavior, inappropriate development, excessive commercialization, and safety hazards. Specifically, for each text in the raw multimodal data, the server can use the text and event prompts as input to a pre-trained event processing model to determine the event information corresponding to the event extracted from the text. The event prompts include a description of the negative event, specifying what constitutes a negative event, and are used to help the event processing model determine whether the extracted event is negative. This event information includes whether the extracted event is negative, the extracted event itself, the corresponding event, location, subject, object, and behavior, among other event elements. For each image in the original multimodal data (each frame in an image or video), the server can determine the corresponding text information (i.e., obtained through OCR recognition technology as described above). This text information, along with event prompts, is used as input to a pre-trained event processing model to determine the event information extracted from the image. Alternatively, if no corresponding text information can be recognized from the image, the event information corresponding to the matching text can be directly used as the event information for the image. This event processing model is a pre-trained large language model. During training, a second text sample and event annotations are first obtained. The event annotations can be BIO (Begin-Inside-Outside) sequence annotations, including whether the event in the second text sample is a negative event, the event itself, the corresponding event, location, subject, object, behavior, and other event elements. The second text sample and event prompts are input into the event processing model to determine the processing result corresponding to the second text sample. Based on the event annotations and processing result, a pre-set event processing loss function is used to determine the second loss. The event processing model is then trained based on this second loss. The event processing loss function mentioned above is preset and can be the cross-entropy loss function or other supervised loss functions. This specification does not impose any specific limitations.

[0091] The target multimodal data obtained after processing by the above data processing strategies includes, in addition to the cleaned original multimodal data, text information corresponding to images in the original multimodal data, matching results of multiple image-text pairs, media level of each data point, sentiment recognition results, and event information. The text information in the target multimodal data can be text, which can be a text segment, text sentence, or article; this specification does not impose specific limitations.

[0092] Based on this, there can be multiple search elements. For ease of explanation, the following description will use search elements. When retrieving data that matches the search elements of the street object, the text information and text in the target multimodal data are used as each search text. From each search text, the search text that matches the search elements of the street object is retrieved and used as the street data. Specifically, this can be achieved by vectorizing the text and then performing vector retrieval. That is, the server can determine the first feature vector corresponding to each search text and the second feature vector corresponding to the search element, calculate the feature similarity between the second feature vector and each first feature vector, and use the search text corresponding to the first feature vector with the highest feature similarity or that reaches the target value as the data that matches the search elements of the street object.

[0093] Of course, when the image cannot be identified as having corresponding text information, the matching relationship between the text matching the image and the search element can be used to determine whether the image is street data. That is, if there is a matching relationship between the text matching the image and the search element, the image is considered as data matching the search element; if there is no matching relationship between the text matching the image and the search element, the image is not considered as data matching the search element.

[0094] S3: Construct a multi-level indicator system for evaluating the effectiveness of neighborhood protection: From the various dimensions of the target multimodal data, determine the dimensions used to evaluate the effectiveness of neighborhood protection as the original evaluation indicators, and based on the original evaluation indicators, use principal component analysis to determine the target evaluation indicators, and construct a multi-level indicator system for evaluating the effectiveness of neighborhood protection.

[0095] In this specification, the server can construct a multi-level indicator system for evaluating the effectiveness of neighborhood protection. Specifically, it determines the dimensions used to evaluate the effectiveness of neighborhood protection from the various dimensions of the target multimodal data as original evaluation indicators. Based on these original evaluation indicators, principal component analysis is used to determine the target evaluation indicators, thus constructing a multi-level indicator system for evaluating the effectiveness of neighborhood protection. Here, each dimension of the target multimodal data refers to the dimension corresponding to the data in the target multimodal data, i.e., the angle, characteristic, or attribute used to describe the data. The data in the target multimodal data corresponds to multiple dimensions. Each original evaluation indicator and its corresponding original indicator value can comprehensively cover the different dimensions of information for each neighborhood object, although there may be some correlation between the original evaluation indicators. The original evaluation indicators must be supported by obtainable actual data; that is, the original indicator values ​​corresponding to the original evaluation indicators must be obtainable from the aforementioned original multimodal data. Furthermore, some relevant theories, such as tourism perceived image and urban image evaluation, are also referenced during the generation process. Therefore, when determining the dimensions used to evaluate the effectiveness of street protection from the various dimensions of the target multimodal data as the original evaluation indicators, the dimensions can be determined manually. Based on these dimensions, original evaluation indicators can be set, and the original indicator values ​​can also be determined manually based on the target multimodal data. Alternatively, a pre-set algorithm can be used to determine the dimensions used to evaluate the effectiveness of street protection from the various dimensions of the target multimodal data, and then original evaluation indicators can be generated based on these dimensions. The number of generated original evaluation indicators can be a pre-set value or the number of dimensions in the target multimodal data; this specification does not specify a particular number. The aforementioned original evaluation indicators are the dimensions used to evaluate the effectiveness of street protection among these multiple dimensions. These dimensions can refer to all dimensions of the target multimodal data, some dimensions, or a combination of at least two dimensions; this specification does not specify a particular number.

[0096] Principal Component Analysis (PCA), as described above, is a classic dimensionality reduction statistical method. Its process follows a "first expand, then simplify" approach: first, it comprehensively considers various factors and indicators, then leverages PCA's data-driven capabilities to extract core elements. This transforms multiple correlated variables (indicators) into a few independent comprehensive indicators, i.e., principal components. Each principal component is a linear combination of the original variables, preserving the information of the original dataset (original multimodal data) to the greatest extent possible. Therefore, in constructing a multi-level indicator system for assessing the effectiveness of historical and cultural district protection, the introduction of PCA primarily addresses the problem of information redundancy among indicators, achieving scientific dimensionality reduction. Since the assessment of historical and cultural districts involves numerous potentially related indicators, such as different types of attention and different dimensions of satisfaction, directly using these indicators leads to information overlap, increases computational complexity, and makes it difficult to determine weights. PCA, however, can extract several core "principal components" from the original indicators. These principal components are independent of each other, reflecting the district's protection effectiveness more concisely and essentially.

[0097] Specifically, in S3 above, when determining the target evaluation indicators and constructing a multi-level indicator system for evaluating the effectiveness of street block protection based on the original evaluation indicators using principal component analysis, the server can first determine the original indicator values ​​corresponding to each original evaluation indicator for each street block object based on the target multimodal data. Each street block object has a corresponding original indicator value for each original evaluation indicator, which is determined from the target multimodal data. The specific determination method can be manual, algorithmic, or model-based; this specification does not specify a particular method. Then, an original data matrix is ​​determined, composed of the original indicator values ​​corresponding to each original evaluation indicator for each street block object. The data elements in this original data matrix represent the original indicator values ​​of the original evaluation indicators for the street block object. The aforementioned original data matrix is... The matrix, This indicates the number of street objects in the target multimodal data. This indicates the number of original evaluation indicators. The target multimodal data includes data corresponding to multiple street objects.

[0098] Since different indicators may have different dimensions and orders of magnitude, the original data matrix needs to be standardized to eliminate their influence and make all indicators comparable. Therefore, the server can standardize each data element in the original data matrix to obtain the target data matrix. This standardization process involves standardizing each original evaluation indicator to a mean of 0 and a standard deviation of 1. Specifically, for each original evaluation indicator, the corresponding data elements in the original data matrix are determined, along with their mean and standard deviation. For each data element of the original evaluation indicator, the difference between the data element and the mean is determined, and then the ratio of this difference to the standard deviation is determined, which is used as the standardized element corresponding to that data element. Based on the standardized elements corresponding to each data element in the original data matrix, the target data matrix is ​​constructed. Each data element in the target data matrix is ​​a standardized element.

[0099] Subsequently, based on the target data matrix, a correlation coefficient matrix is ​​calculated using principal component analysis (PCA), and eigenvalue decomposition (EVD) is performed on the correlation coefficient matrix to obtain the eigenvalues ​​and eigenvectors corresponding to each principal component. This correlation coefficient matrix is ​​also a covariance matrix, used to characterize the linear correlation between the original evaluation indicators. Specifically, this correlation coefficient matrix can be calculated using the number of objects in the block and the target data matrix, and it can be... The correlation coefficient matrix can be a matrix. , This represents the target data matrix. The eigenvalues ​​described above characterize the variance explained by the principal components, i.e., the amount of information contained in the original multimodal data by the principal components. The larger the eigenvalue, the more important the corresponding principal component. The eigenvectors characterize the direction of the principal components, i.e., the linear relationship between the original evaluation indicators, and have a length of [missing information]. The principal component is a comprehensive scalar, and it is the projection score of the original index value corresponding to each original evaluation index of all block objects in the composition direction.

[0100] Next, based on the eigenvalues ​​of each principal component, the principal components are screened to obtain the target principal components. This screening method can be based on the eigenvalue rule (Kaiser's criterion) or on the cumulative contribution rate criterion. Specifically, the server can use the Kaiser's criterion to determine the principal components whose eigenvalues ​​are greater than a first threshold, and use these as target principal components. This first threshold can be pre-set, such as 0.9, or other values; this specification does not specify a particular value. The server can also calculate the variance contribution rate of each principal component based on its eigenvalues, which is the sum of the eigenvalues ​​of each principal component and the total eigenvalue (the sum of the eigenvalues ​​of all principal components). Based on the variance contribution rate of each principal component, the cumulative contribution rate of each principal component is calculated using the following formula:

[0101]

[0102] in, Indicates the first The cumulative contribution rate of each principal component Indicates the first The eigenvalues ​​of the principal components are given. These principal components can be arranged in descending order of their eigenvalues ​​to obtain a principal component sequence. Then, the aforementioned principal component sequence... The principal component is actually the first one in the principal component sequence. The principal component, located at the _ , is the _ . The eigenvalues ​​of the principal components preceding the first principal component are all higher than those of the first principal component. The eigenvalues ​​of the principal components are large.

[0103] The principal components whose cumulative contribution rate is greater than a second threshold are identified and designated as the first principal components. The principal components preceding the first principal components in the principal component sequence are identified and designated as the second principal components. The first and second principal components are then used as the target principal components. The second threshold can be preset; it can be 0.85, or any other value, which is not specifically limited in this specification. The principal component sequence is obtained by arranging the principal components in descending order of their eigenvalues. Assuming the first principal component is the [missing value] in the principal component sequence... One principal component, then The number of principal components retained, and the number of target principal components. And the first part of the principal component sequence One principal component. The target principal components are obtained through the above screening method, retaining those with high cumulative variance contribution rates, which can represent most of the information in the original multimodal data.

[0104] The server then identifies the elements in the eigenvector corresponding to each target principal component whose loadings reach a specified threshold, and uses these as target elements. The server determines the original evaluation index corresponding to each target element and uses it as the target evaluation index. Based on these target evaluation indices, a multi-level index system for evaluating the effectiveness of street protection is constructed. The aforementioned loadings characterize the correlation strength between the original evaluation indices and the principal components. These loadings are element values ​​in the eigenvector, and the specified threshold is pre-set; this threshold can be 0.8, or other values, which are not specifically limited in this specification. The eigenvector includes multiple elements, each with a corresponding original evaluation index.

[0105] When constructing a multi-level indicator system for evaluating the effectiveness of street protection based on various target evaluation indicators, the server can use each target evaluation indicator as a tertiary evaluation indicator, then construct secondary evaluation indicators based on each tertiary evaluation indicator, and then construct primary evaluation indicators based on each secondary evaluation indicator. Finally, based on each tertiary evaluation indicator, each secondary evaluation indicator, and each primary evaluation indicator, a multi-level indicator system for evaluating the effectiveness of street protection is constructed. The process of constructing the secondary and primary evaluation indicators can be done manually or automatically by a pre-set algorithm; this specification does not impose specific limitations.

[0106] The aforementioned multi-level indicator system for assessing the effectiveness of street block protection is used to evaluate the protection effectiveness of each street block (historical and cultural street block). This system includes primary, secondary, and tertiary assessment indicators. Each primary indicator includes a first number of secondary indicators, and each secondary indicator includes a second number of tertiary indicators. The first and second numbers are pre-set, and the number of secondary indicators (i.e., the first number) under each primary indicator can be the same or different; for example, the first number can be set to 3. The number of tertiary indicators (i.e., the second number) under each secondary indicator can also be the same or different, and can be set according to needs. The number of primary indicators (i.e., the third number) included in the multi-level indicator system for assessing the effectiveness of street block protection can also be pre-set, for example, 2. Specifically, as shown... Figure 2 As shown, Figure 2 This is a schematic diagram of a multi-level indicator system for evaluating the effectiveness of neighborhood protection, as provided in this specification. Figure 2 This is merely an example of a multi-level indicator system for evaluating the effectiveness of neighborhood protection, and this specification is not restrictive. The primary evaluation indicators in the aforementioned multi-level indicator system for evaluating neighborhood protection effectiveness may be influence and reputation. Secondary evaluation indicators under influence may include public attention, media attention, and protection investment. Secondary evaluation indicators under reputation may include public satisfaction, media recognition, and negative warning level.

[0107] The three-tiered evaluation indicators under the aforementioned public attention level may include public attention on the first UGC platform, public attention on the second UGC platform, public attention on third-party platforms, and public attention on short video platforms. The description of public attention on the first UGC platform can be the number of posts related to relevant historical and cultural blocks on the first UGC platform. The description of public attention on the second UGC platform can be the number of posts related to relevant historical and cultural blocks on the second UGC platform. The description of public attention on third-party platforms can be the number of posts related to relevant historical and cultural blocks on self-media accounts on various third-party platforms. The description of public attention on short video platforms can be the number of posts related to relevant historical and cultural blocks on short videos on various short video platforms. The aforementioned UGC platforms are user-generated content platforms. The first and second UGC platforms are different user-generated content platforms. The aforementioned third-party platforms are platforms that can run self-media accounts. The aforementioned short video platforms are network service platforms that support users uploading and playing video content with a duration less than a specified length. The specified duration mentioned above is the duration limited by the short video platform. The duration limit may be the same or different for different short video platforms. It should be noted that this specification does not limit the specific platforms corresponding to the first UGC platform, the second UGC platform, third-party platforms, self-media accounts, and short video platforms mentioned above, and can be determined according to the actual situation.

[0108] The three-tiered evaluation indicators for media attention mentioned above may include core media attention, local and industry media attention, and self-media attention. The description of core media attention can be the total amount of reports about relevant historical and cultural blocks on multiple channels across the main accounts of multiple core media outlets. These multiple core media outlets include, but are not limited to, various news platforms. These multiple channels may include websites, mobile applications, UGC platforms, etc. The description of local and industry media attention can be the total amount of reports about relevant historical and cultural blocks on multiple channels by media other than core media. These other media outlets may be sourced from the "List of Internet News Information Source Units". The description of self-media attention can be the total amount of reports about relevant historical and cultural blocks by other self-media outlets on multiple channels.

[0109] The three-tiered assessment indicators under the aforementioned level of protection investment may include policy visibility, the number of articles published by local management agencies concerning historical and cultural blocks, and the number of activities published by local management agencies concerning historical and cultural blocks. The description of policy visibility may include, but is not limited to, the number of special policy documents in areas such as urban and rural construction and cultural heritage protection, and the number of documents in annual economic and social development reports that mention keywords such as "historical and cultural city," "block," and "historical building" within historical and cultural blocks. The description of the number of articles published by local management agencies concerning historical and cultural blocks may include, the number of news reports published by comprehensive management entities legally responsible for professional public affairs management in areas such as urban renewal, cultural heritage protection, restoration, and revitalization, covering relevant historical and cultural blocks, across multiple channels. The description of the number of activities published by local management agencies concerning historical and cultural blocks may include, the number of news reports published by comprehensive management entities legally responsible for professional public affairs management in areas such as urban renewal, cultural heritage protection, restoration, and revitalization, covering activities within historical and cultural blocks, across multiple channels.

[0110] The three-tiered evaluation indicators for public satisfaction mentioned above may include positive public opinion and positive public interaction. Positive public opinion can be described as the number of articles positively evaluating historical and cultural districts from various online data collected from different channels. Positive public interaction can be described as the ratio of the sum of reposts, likes, and comments on positive articles to the total number of positive articles.

[0111] The three-tiered evaluation indicators under the aforementioned media recognition can include positive media coverage and positive coverage from local management agencies. Positive media coverage can be described as the number of articles in media outlets that positively evaluate historical and cultural blocks. Positive coverage from local management agencies can be described as the number of articles in departments legally responsible for public affairs management that positively evaluate historical and cultural blocks.

[0112] The three-tiered assessment indicators under the aforementioned negative warning level may include the volume of negative voices concerning the historical and cultural district, the number of negative events, and the number of key sources mentioning the district. The description of the volume of negative voices concerning the historical and cultural district can be the number of articles negatively evaluating the historical and cultural district from various internet channels. The description of the number of negative events can be the number of negative events related to the preservation, protection, and revitalization of historical heritage in the historical and cultural district. The description of the number of key sources mentioning the district can be the number of negative articles from key sources (i.e., news organizations marked as important) from various internet channels concerning the historical and cultural district.

[0113] S4: Evaluate the effectiveness of the protection of historical and cultural blocks: Based on the block data and the multi-level indicator system for evaluating the effectiveness of block protection, evaluate the effectiveness of the protection of each block.

[0114] In this specification, the server can assess the effectiveness of historical and cultural district preservation. Specifically, it evaluates the preservation effectiveness of each district based on district data and a multi-level indicator system for assessing preservation effectiveness. The specific steps are as follows:

[0115] S41: Based on the street data of each street object, determine the indicator value of each third-level evaluation indicator corresponding to each street object.

[0116] S42: Construct the original matrix of tertiary evaluation indicators based on the indicator values ​​of each tertiary evaluation indicator for each block object.

[0117] S43: Based on the original matrix of the three-level evaluation indicators, determine the target indicator weight corresponding to each evaluation indicator in the multi-level indicator system for evaluating the effectiveness of street protection.

[0118] S44: Determine the protection effectiveness index for each block based on the target indicator weights and the original matrix of the three-level evaluation indicators for each evaluation indicator.

[0119] Specifically, in step S41 above, when determining the indicator value of each third-level evaluation indicator for each street object based on its street data, the server can, for each street object, extract the data corresponding to each third-level evaluation indicator from its street data according to the description of each third-level evaluation indicator, and use this data as the indicator value for each third-level evaluation indicator of that street object. The aforementioned description describes how the indicator values ​​corresponding to the third-level evaluation indicators are obtained, or what they represent. The server can automatically obtain the indicator values ​​corresponding to the third-level evaluation indicators from the street data using a preset algorithm or model based on the description of each third-level evaluation indicator.

[0120] The original matrix of the above-mentioned three-level evaluation indicators can be directly constructed from the indicator values ​​of each three-level evaluation indicator for each block object. Specifically, taking n block objects and m three-level evaluation indicators as an example, the original matrix of the above-mentioned three-level evaluation indicators can be expressed by the following formula:

[0121]

[0122] in, This represents the original matrix of the three-level evaluation indicators. ~ This represents the values ​​of the m tertiary evaluation indicators for the first block object, in terms of... For example, This represents the value of the first tertiary evaluation indicator for the first block object. ~ This represents the value of m third-level evaluation indicators for the nth block object, in... For example, This represents the value of the m-th level 3 evaluation indicator for the n-th block object.

[0123] In S43 above, when determining the target indicator weight corresponding to each evaluation indicator in the multi-level indicator system for assessing the effectiveness of street protection based on the original matrix of the three-level indicators, the server can execute the following steps:

[0124] S431: Use the three-level evaluation indicators as the evaluation indicators to be processed, and use the original matrix of the three-level evaluation indicators as the original matrix of the evaluation indicators to be processed.

[0125] S432: Regularize the original matrix of the evaluation indicators to be processed to obtain the standardized matrix of the evaluation indicators to be processed.

[0126] S433: Based on the standardized matrix of the evaluation indicators to be processed, the entropy value corresponding to each evaluation indicator to be processed is calculated using the entropy method.

[0127] S434: Determine the difference coefficient for each evaluation indicator to be processed based on the entropy value corresponding to each evaluation indicator to be processed.

[0128] S435: Normalize the difference coefficients of each evaluation indicator to be processed under the same superior evaluation indicator to obtain the target indicator weights of each evaluation indicator to be processed under the same superior evaluation indicator.

[0129] S436: When the evaluation indicator to be processed is not a first-level evaluation indicator, determine the indicator value corresponding to the superior evaluation indicator of each evaluation indicator to be processed for each block object based on the target indicator weight and the standardized matrix of the evaluation indicator to be processed, and construct the original matrix of the superior evaluation indicator.

[0130] S437: Reuse the superior evaluation indicators as the evaluation indicators to be processed, and reuse the original matrix of the superior evaluation indicators as the original matrix of the evaluation indicators to be processed. Repeat steps S432 to S437 until the evaluation indicators to be processed are first-level evaluation indicators, and obtain the target indicator weights corresponding to each evaluation indicator in the multi-level indicator system for evaluating the effectiveness of street protection.

[0131] In step S432 above, when regularizing the original matrix of evaluation indicators to obtain the standardized matrix, the level of the evaluation indicators needs to be considered. Different levels of evaluation indicators require different regularization methods. Specifically, when the original matrix of evaluation indicators is a level 3 matrix, the server can use a preset first regularization formula to regularize the indicator values ​​corresponding to positive evaluation indicators in the original matrix, and a preset second regularization formula to regularize the indicator values ​​corresponding to negative evaluation indicators in the original matrix, to obtain the standardized matrix of evaluation indicators. When the original matrix of evaluation indicators is not a level 3 matrix, the first regularization formula is used to regularize the indicator values ​​corresponding to all evaluation indicators in the original matrix.

[0132] The elements in the standardized matrix of the above-mentioned evaluation indicators are all regularized indicator values. Both the positive and negative evaluation indicators are pre-set and are all level 3 evaluation indicators. The negative evaluation indicators may include all level 3 evaluation indicators under the negative warning level, and the positive evaluation indicators are level 3 evaluation indicators other than the negative evaluation indicators. The first and second regularization formulas are pre-set formulas. The first regularization formula is as follows:

[0133]

[0134] The second regularization formula mentioned above is as follows:

[0135]

[0136] in, This represents the first element in the original matrix of evaluation indicators for treatment. The first block object The index values ​​are obtained by regularizing the index values ​​corresponding to the evaluation indicators to be processed. This represents the first element in the original matrix of evaluation indicators to be processed. The first block object The indicator values ​​corresponding to the evaluation indicators to be processed. This represents the first element in the original matrix of evaluation indicators to be processed. The minimum indicator value corresponding to each evaluation indicator to be processed This represents the first element in the original matrix of evaluation indicators to be processed. The maximum value corresponding to each pending evaluation indicator. (The above) , , The original matrix of evaluation indicators has different values ​​at different levels, then the corresponding values ​​are... They are also different. When the evaluation indicator to be processed is a level three evaluation indicator, the above... For the above In , For the above The first in The minimum indicator value corresponding to each of the three-level evaluation indicators, i.e. , For the above The first in The maximum value corresponding to each of the three-level evaluation indicators, i.e. , The maximum value is m. When the evaluation indicator to be processed is a secondary evaluation indicator, the above... For the following In , For the following The first in The minimum indicator value corresponding to each secondary evaluation indicator, i.e. , For the following The first in The maximum value corresponding to each secondary evaluation indicator, i.e. , The maximum value is p. When the evaluation indicator to be processed is a primary evaluation indicator, the above... For the following In , For the following The first in The minimum indicator value corresponding to each primary evaluation indicator, i.e. , For the following The first in The maximum value corresponding to each primary evaluation indicator, i.e. , The maximum value is q. The maximum value is n, which is the number of objects in the block. This is the original matrix of secondary evaluation indicators. This is the original matrix of primary evaluation indicators.

[0137] The original matrix of the above secondary evaluation indicators can be represented by the following formula:

[0138]

[0139] in, This represents the original matrix of secondary evaluation indicators. ~ This represents the index values ​​of p secondary evaluation indicators for the first block object, in... For example, This represents the value of the first secondary evaluation indicator for the first block object. ~ This represents the index values ​​of p secondary evaluation indicators for the nth block object, in... For example, This represents the value of the p-th secondary evaluation indicator for the n-th block object.

[0140] The original matrix of the above primary evaluation indicators can be represented by the following formula:

[0141]

[0142] in, This represents the original matrix of primary evaluation indicators. ~ This represents the index values ​​of the q primary evaluation indicators for the first block object, in... For example, This represents the value of the first primary evaluation indicator for the first block object. ~ This represents the value of the q primary evaluation indicators for the nth block object, in... For example, This represents the value of the q-th primary evaluation indicator for the nth block object.

[0143] Different levels of evaluation indicators correspond to different standardized evaluation indicator matrices. If the evaluation indicator to be processed is a level 3 evaluation indicator, then the standardized evaluation indicator matrix to be processed is the level 3 evaluation indicator standardized matrix, that is... Specifically, the following formula can be used:

[0144]

[0145] in, ~ for ~ The corresponding regularized index value.

[0146] If the evaluation indicator to be processed is a secondary evaluation indicator, then the standardized matrix of the evaluation indicator to be processed is the standardized matrix of the secondary evaluation indicator, that is... Specifically, the following formula can be used:

[0147]

[0148] in, ~ for ~ The corresponding regularized index value.

[0149] If the evaluation indicator to be processed is a primary evaluation indicator, then the standardized matrix of the evaluation indicator to be processed is the standardized matrix of the primary evaluation indicator, that is... Specifically, the following formula can be used:

[0150]

[0151] in, ~ for ~ The corresponding regularized index value.

[0152] In S433 above, when calculating the entropy value corresponding to each evaluation indicator using the entropy method based on the standardized matrix of the evaluation indicators to be processed, the first entropy value calculation formula can be used to calculate the proportion of each block object on each evaluation indicator to be processed based on the standardized matrix of the evaluation indicators to be processed. Then, based on the proportion of each block object on each evaluation indicator to be processed, the second entropy value calculation formula can be used to calculate the entropy value corresponding to each evaluation indicator to be processed. The first entropy value calculation formula can be:

[0153]

[0154] in, Indicates the first The block object in the first The weighting of each pending evaluation indicator This represents the first element in the standardized matrix of the evaluation indicators to be processed. The first block object The indicator values ​​corresponding to the evaluation indicators to be processed (i.e., the regularized indicator values), where n represents the number of objects in the block. If the above evaluation indicators to be processed are level three evaluation indicators, then for ,and Indicates the first The block object in the first The weighting of each of the three-level evaluation indicators. If the above-mentioned evaluation indicator to be processed is a second-level evaluation indicator, then... for ,and Indicates the first The block object in the first The weighting of each secondary evaluation indicator. If the above-mentioned evaluation indicator to be processed is a primary evaluation indicator, then... for ,and Indicates the first The block object in the first The weighting of each primary evaluation indicator.

[0155] The formula for calculating the second entropy value mentioned above can be:

[0156]

[0157] in, Indicates the first The entropy value corresponding to each evaluation indicator to be processed, which can be a third-level evaluation indicator, a second-level evaluation indicator, or a first-level evaluation indicator.

[0158] When calculating the difference coefficient in S434 above, the difference coefficient for each evaluation indicator to be processed can be determined using the difference coefficient calculation formula based on the entropy value corresponding to each evaluation indicator to be processed. The difference coefficient calculation formula is as follows:

[0159]

[0160] in, Indicates the first The difference coefficients corresponding to the evaluation indicators to be processed, which can be tertiary evaluation indicators, secondary evaluation indicators, or primary evaluation indicators.

[0161] In step S435 above, when calculating the target indicator weights corresponding to the evaluation indicators to be processed, the server can group the evaluation indicators to be processed according to their corresponding superior evaluation indicators to obtain indicator groups. The evaluation indicators to be processed in each indicator group are under the same superior evaluation indicator, meaning they belong to the same superior evaluation indicator. The difference coefficients corresponding to the evaluation indicators to be processed within each indicator group are normalized to determine the target indicator weights for each evaluation indicator to be processed within each indicator group. Specifically, when normalizing the difference coefficients corresponding to the evaluation indicators to be processed within each indicator group, the following normalization formula can be used to calculate the corresponding indicator weights for each evaluation indicator to be processed within each indicator group:

[0162]

[0163] in, Indicates the first The first of the indicator groups The coefficient of difference for each evaluation indicator to be processed varies depending on the specific indicator group included. Indicates the first The number of evaluation indicators to be processed included in each indicator group. The number of evaluation indicators to be processed included in different indicator groups may be the same or different. Represents the calculated first The first of the indicator groups Each evaluation indicator to be processed corresponds to an indicator weight, which can be a tertiary evaluation indicator, a secondary evaluation indicator, or a primary evaluation indicator.

[0164] The indicator weights corresponding to each evaluation indicator to be processed in each indicator group, calculated using the normalization formula above, can be directly used as the target indicator weights corresponding to each evaluation indicator to be processed in each indicator group.

[0165] Furthermore, since assessment indicators with high importance and low data volatility may have low corresponding indicator weights, subjective weights can be assigned to these indicators using the Delphi method. Among the aforementioned assessment indicators to be processed, those with high importance levels and low data fluctuations can be designated as the primary assessment indicators. Importance level characterizes the degree of influence of the assessed indicator on the protection effectiveness assessment; the higher the importance level, the greater the influence, and vice versa. Data fluctuation characterizes the stability of the indicator value corresponding to the assessed indicator; the smaller the data fluctuation, the more stable the indicator value, and vice versa. This primary assessment indicator is pre-set, i.e., manually set, but can also be set (determined) based on the importance level and data fluctuation of each assessed indicator to be processed. The importance level of each assessed indicator to be processed can be manually set, for example, by assigning an importance level to each assessed indicator through expert scoring, or by determining the importance level of each assessed indicator through a model, such as using random forests to output the feature importance of each assessed indicator and use it as the importance level. This specification does not specify any particular limitation. The aforementioned data fluctuation can be determined based on the indicator values ​​corresponding to the assessed indicators for each block object, specifically the variance or standard deviation of the indicator values ​​corresponding to the assessed indicators for each block object. This specification does not specify any particular limitation. When determining the first evaluation indicator, indicators with an importance level greater than a first specific threshold and data fluctuation less than a second specific threshold can be used as the first evaluation indicator. The first and second specific thresholds are preset. Alternatively, a score can be determined for each indicator based on its importance level and data fluctuation, and the indicator with the highest score can be used as the first evaluation indicator. Higher importance levels and lower data fluctuations result in higher scores for the corresponding indicator, and vice versa. This first evaluation indicator can be a secondary evaluation indicator, specifically a negative warning level.

[0166] Based on this, in S435 above, when the evaluation indicator to be processed is a secondary evaluation indicator, the difference coefficients corresponding to each evaluation indicator to be processed under the same superior evaluation indicator are normalized to determine the initial indicator weights corresponding to each evaluation indicator to be processed under the same superior evaluation indicator. Based on the Delphi method, the initial indicator weights corresponding to the first evaluation indicator among the evaluation indicators to be processed are adjusted to obtain the target indicator weights corresponding to the first evaluation indicator. Evaluation indicators to be processed that belong to the same superior evaluation indicator as the first evaluation indicator are determined and used as related evaluation indicators. Based on the target indicator weights corresponding to the first evaluation indicator, the initial indicator weights corresponding to the related evaluation indicators are adjusted to obtain the target indicator weights corresponding to the related evaluation indicators. The initial indicator weights corresponding to the other evaluation indicators to be processed besides the first evaluation indicator and the related evaluation indicators are used as the target indicator weights.

[0167] The process of determining the initial indicator weights for each evaluation indicator under the same superior evaluation indicator is consistent with the process of first grouping and then calculating the weights within each group based on a normalization formula; the specific method will not be elaborated here. When adjusting the initial indicator weights corresponding to the first evaluation indicator among the evaluation indicators to obtain the target indicator weights corresponding to the first evaluation indicator based on the Delphi method, the subjective weights corresponding to the first evaluation indicator can be obtained first. These subjective weights are assigned to the first evaluation indicator by domain experts based on their experience. The subjective weights are then directly used as the target indicator weights corresponding to the first evaluation indicator. The related evaluation indicator is an evaluation indicator to be processed that belongs to the same superior evaluation indicator as the first evaluation indicator. This related evaluation indicator can be a secondary evaluation indicator, and the superior evaluation indicator of the first evaluation indicator is a primary evaluation indicator. When adjusting the initial indicator weights corresponding to the related evaluation indicators based on the target indicator weights corresponding to the first evaluation indicator to obtain the target indicator weights corresponding to the related evaluation indicators, the following modified calculation formula can be used to adjust the initial indicator weights corresponding to the related evaluation indicators:

[0168]

[0169] in, Indicates the first The initial indicator weights corresponding to each correlation evaluation indicator This represents the subjective weight, that is, the target indicator weight corresponding to the first evaluation indicator. Indicates the first The adjusted indicator weights corresponding to each related evaluation indicator, i.e., the target indicator weights. Indicates the number of related evaluation indicators. This represents the sum of the initial indicator weights for all related evaluation indicators.

[0170] It should be noted that when the evaluation indicator to be processed is a first-level evaluation indicator, the superior evaluation indicator of the evaluation indicator to be processed is empty, and all evaluation indicators to be processed belong to the same superior evaluation indicator.

[0171] In S436 above, the server can first determine whether the evaluation indicator to be processed is a primary evaluation indicator. If not, when the evaluation indicator to be processed is not a primary evaluation indicator, it can weight the indicator value corresponding to each evaluation indicator of each block object in the standardized matrix of evaluation indicators to be processed according to the target indicator weight corresponding to each evaluation indicator to be processed, to obtain the weighted indicator value corresponding to each evaluation indicator to be processed for each block object. It then determines the sum of the weighted indicator values ​​corresponding to the evaluation indicators to be processed belonging to the same superior evaluation indicator for each block object, and based on the determined sum, determines the indicator value corresponding to each superior evaluation indicator for each block object, and determines the original matrix of superior evaluation indicators composed of the indicator values ​​corresponding to each superior evaluation indicator for each block object. The weighted indicator values ​​can be obtained through... Calculations show that Indicates the first The target indicator weights for each evaluation indicator to be processed are determined, since the evaluation indicators to be processed are not primary evaluation indicators. It can represent the first The target indicator weights for each secondary or tertiary evaluation indicator. This represents the first element in the standardized matrix of secondary or tertiary evaluation indicators. The first block object The indicator values ​​corresponding to each secondary or tertiary evaluation indicator (i.e., the regularized indicator values), namely or The original matrix of higher-level evaluation indicators can be either the original matrix of secondary evaluation indicators or the original matrix of primary evaluation indicators.

[0172] The above-mentioned method for determining the sum of weighted index values ​​corresponding to the unprocessed evaluation indicators belonging to the same superior evaluation indicator for each block object, and determining the index value corresponding to each superior evaluation indicator for each block object based on the determined sum, can first group the unprocessed evaluation indicators according to the superior evaluation indicator corresponding to each unprocessed evaluation indicator, resulting in index groups. The unprocessed evaluation indicators in each index group are under the same superior evaluation indicator, that is, they belong to the same superior evaluation indicator, and each index group has a corresponding superior evaluation indicator. The number of index groups corresponds to the number of superior evaluation indicators. Then, for each block object, and for each index group of that block object, the sum of weighted index values ​​corresponding to the unprocessed evaluation indicators included in each index group of that block object is determined, and this sum is used as the index value corresponding to the superior evaluation indicator corresponding to each index group of that block object. This process determines the index value corresponding to each superior evaluation indicator for each block object.

[0173] Specifically, taking the following example: the evaluation indicators to be processed are tertiary evaluation indicators, the superior evaluation indicators are secondary evaluation indicators, the number of indicator groups (superior evaluation indicators) is 6, and the number of evaluation indicators to be processed is 17. Assuming that the 1st to 4th evaluation indicators to be processed belong to indicator group 1, the 5th to 7th belong to indicator group 2, the 8th to 10th belong to indicator group 3, the 11th to 12th belong to indicator group 4, the 13th to 14th belong to indicator group 5, and the 15th to 17th belong to indicator group 6, the indicator value corresponding to each superior evaluation indicator for each block object can be determined using the following formula:

[0174]

[0175]

[0176]

[0177]

[0178]

[0179]

[0180] in, ~ Indicates the first The indicator values ​​corresponding to the 6 indicator groups (superior evaluation indicators) for each block object, Indicates the first The first block object The indicator values ​​(i.e., regularized indicator values) corresponding to the three-level evaluation indicators are given in the above formula. Indicates the first The target indicator weights for each of the three-level evaluation indicators.

[0181] Taking the following example, where the evaluation indicators to be processed are secondary evaluation indicators, the superior evaluation indicators are primary evaluation indicators, the number of indicator groups (superior evaluation indicators) is 2, and the number of evaluation indicators to be processed is 6, assuming that the 1st to 3rd evaluation indicators to be processed belong to the 1st indicator group, and the 4th to 6th evaluation indicators to be processed belong to the 2nd indicator group, the following formula can be used to determine the indicator value corresponding to each superior evaluation indicator for each block object:

[0182]

[0183]

[0184] in, ~ Indicates the first The indicator values ​​corresponding to the two indicator groups (superior evaluation indicators) for each block object. Indicates the first The first block object The indicator values ​​(i.e., the regularized indicator values) corresponding to each secondary evaluation indicator, in the above formula Indicates the first The target indicator weights for each secondary evaluation indicator.

[0185] In step S44 above, when determining the protection effectiveness index for each block object based on the target indicator weights and the original matrix of the three-level evaluation indicators for each evaluation indicator, the server can regularize the original matrix of the three-level evaluation indicators to obtain a standardized matrix of the three-level evaluation indicators. Based on the standardized matrix of the three-level evaluation indicators, the third indicator value corresponding to each three-level evaluation indicator for each block object is determined. Using the target indicator weights of each three-level evaluation indicator, the third indicator value of each three-level evaluation indicator for each block object is weighted to obtain the third weighted value of each three-level evaluation indicator for each block object. Based on the third weighted value of each three-level evaluation indicator for each block object, the second indicator value of each two-level evaluation indicator for each block object is determined. Using the target indicator weights of each two-level evaluation indicator, the second indicator value of each two-level evaluation indicator for each block object is weighted to obtain the second weighted value of each two-level evaluation indicator for each block object. Based on the second weighted value of each two-level evaluation indicator for each block object, the first indicator value of each one-level evaluation indicator for each block object is determined. By employing the target indicator weights of each primary evaluation indicator, the first indicator value of each primary evaluation indicator for each block is weighted to obtain the first weighted value of each primary evaluation indicator for each block. The sum of the first weighted values ​​of each primary evaluation indicator for each block is determined and used as the corresponding protection effectiveness index for each block.

[0186] In this context, the third indicator value corresponding to each third-level evaluation indicator for each of the aforementioned block objects is an element in the standardized matrix of the third-level evaluation indicators. The aforementioned third weighted value is the value obtained by weighting the third indicator value with the target indicator weights of the third-level evaluation indicators. The second indicator value for each second-level evaluation indicator for each of the aforementioned block objects is an element in the original matrix of the second-level evaluation indicators. Specifically, for each block object, the sum of the third weighted values ​​of the third evaluation indicators belonging to each second-level evaluation indicator for that block object can be determined and used as the indicator value of each second-level evaluation indicator for that block object, i.e., the second indicator value. The aforementioned second weighted value is the value obtained by weighting the second indicator value with the target indicator weights of the second-level evaluation indicators. When determining the first indicator value for each first-level evaluation indicator for each block object, the sum of the second weighted values ​​of the second-level evaluation indicators belonging to each first-level evaluation indicator for that block object can be determined and used as the indicator value of each first-level evaluation indicator for that block object, i.e., the first indicator value. The first weighted value is the value obtained by weighting the first indicator value with the target indicator weights of the first-level evaluation indicators. The protection effectiveness index for each block object can be the sum of the first weighted values ​​of each first-level evaluation indicator for each block object.

[0187] Taking 17 tertiary evaluation indicators, 6 secondary evaluation indicators, and 2 primary evaluation indicators as an example, assuming that the 1st to 4th tertiary evaluation indicators belong to the 1st secondary evaluation indicator, the 5th to 7th tertiary evaluation indicators belong to the 2nd secondary evaluation indicator, the 8th to 10th tertiary evaluation indicators belong to the 3rd secondary evaluation indicator, the 11th to 12th tertiary evaluation indicators belong to the 4th secondary evaluation indicator, the 13th to 14th tertiary evaluation indicators belong to the 5th secondary evaluation indicator, and the 15th to 17th tertiary evaluation indicators belong to the 6th secondary evaluation indicator, the protection effectiveness index for each block can be calculated using the following protection effectiveness index:

[0188]

[0189] in, Indicates the first The protection effectiveness index of each neighborhood object ~ This indicates the target indicator weights corresponding to the first and second primary evaluation indicators. ~ This indicates the target indicator weights corresponding to the 1st to 6th secondary evaluation indicators. Indicates the first The target indicator weights corresponding to each of the three-level evaluation indicators. Indicates the first The first block object The regularized values ​​of each of the three-level evaluation indicators.

[0190] The aforementioned protection effectiveness index characterizes the protection effectiveness of the block object (historical and cultural block), that is, the evaluation result. In other words, the protection effectiveness index can be directly used as the evaluation result of the block object. The larger the protection effectiveness index, the better the protection effectiveness of the block object. Conversely, the smaller the protection effectiveness index, the worse the protection effectiveness of the block object.

[0191] In some embodiments of this specification, to ensure the accuracy of the street elements in the element pool generated by the large-scale street element generation model, the server can perform element cleaning on the street elements in the element pool generated by the large-scale street element generation model, and use a knowledge verification model to further verify the cleaned street elements to ensure the comprehensiveness, accuracy, and consistency of the generated street elements. Specifically, in S1 above, the server can obtain a list of street names and task prompts. The list of street names and task prompts are input into the large-scale street element generation model. The large-scale street element generation model is enhanced by searching through pre-built knowledge bases. The large-scale street element generation model generates street elements for each street object in the list of street names, and divides the street elements of each street object based on street name, entity elements, spatial elements, and social element categories. The results of the division of each street object are combined into an initial element pool for each street object, which is used as the output of the large-scale street element generation model. The street elements in the initial element pool of each street object are cleaned. The street elements in each initial element pool after cleaning are verified by the knowledge verification model to obtain the target element pool for each street object.

[0192] Specifically, when cleaning the street elements in the initial element pool for each street object, a pre-set knowledge engineering set can be used to clean the street elements in the initial element pool for each street object. This standardizes the street elements in the initial element pool for each street object, resolving issues such as fragmentation, redundancy, and semantic ambiguity of the element pool information. The aforementioned knowledge engineering set includes various knowledge engineering techniques, specifically entity extraction, entity linking, conflict detection, and knowledge completion, and may also include relation extraction. Each initial element pool undergoes processing through these multiple knowledge engineering processes. The processing of the initial element pool by these knowledge engineering techniques is pipelined, meaning that the element cleaning of the initial element pool is performed in the order of entity extraction, entity linking, conflict detection, and knowledge completion. Specifically, as follows... Figure 3 As shown, Figure 3 This is a schematic diagram illustrating a process for generating street features as provided in this specification. Figure 3 The first step is to compile a list of street names (i.e.) Figure 3 The input ("Input: List of Street Names") and task prompts are used to input the street element generation model. Based on pre-built knowledge bases, the street element generation model generates an initial element pool for each street object in the list of street names. Figure 3The "initial generation" and "initial element pool" in the knowledge base include a geographic information database, a neighborhood encyclopedia, a historical document database, and a conservation planning knowledge base. The initial element pool includes neighborhood names, entity elements, spatial elements, and social elements. Then, a pre-defined knowledge engineering set is used to clean the neighborhood elements in each initial element pool. This knowledge engineering set includes entity extraction, entity linking, conflict detection, and knowledge completion. A large-scale knowledge verification model is then used to verify the cleaned neighborhood elements in each initial element pool to obtain the target element pool for each neighborhood object. Figure 3 The output is "Pool of each target element".

[0193] The entity extraction described above is used to extract entities of a preset type from the feature pool. These entities can include buildings, streets, etc. (from entity features), people, communities, events, etc. (from social features), and can also include time and location. The specific types of entities extracted can be preset. Entity extraction can be implemented using a pre-trained model or a pre-set algorithm. Taking a pre-trained entity extraction model as an example, each initial feature pool is input into the pre-trained entity extraction model, and the model extracts entities from each initial feature pool. The entity extraction model can be a pre-trained BERT model. During training, a sample feature pool and entity labels are first obtained. The sample feature pool is then input into the entity extraction model to obtain output entities. Based on the output entities and entity labels, a preset entity loss function is used to calculate the entity loss. The entity extraction model is then trained based on the entity loss. The entity loss function is preset and can be a cross-entropy loss function or other supervised loss functions; this description does not specify a particular limitation.

[0194] The aforementioned entity chain refers to the technique of linking extracted entities to unique standard entities in a knowledge base by combining street elements in the element pool and various knowledge bases. These standard entities can also be extracted using the aforementioned model (e.g., an entity extraction model) or algorithm. Specifically, for each extracted first entity (i.e., extracted from the element pool), a matching standard entity is determined from the standard entities in each knowledge base, and this matching standard entity is used as the associated standard entity. However, if multiple matching standard entities are determined, meaning one first entity is associated with multiple standard entities (e.g., the extracted entity is "A Street," but the standard entities include "A Street in Region Q" and "A Street in Region W"), then determining the matching standard entity for "A Street" will result in two outcomes: "A Street in Region Q" and "A Street in Region W." At this point, a multi-factor fusion model can be used to comprehensively improve disambiguation accuracy by integrating information from different dimensions such as entity type prediction, matching context, and vector similarity calculation. This associates the extracted entities with the unique and correct standard entity in the knowledge base, identifying "same name, different meaning" and "synonymous" issues. Specifically, the multiple standard entities that match the first entity identified above can be used as candidate entities. The multi-factor fusion model, based on the knowledge base and initial element pool, determines the target entity that matches the first entity from among the candidate entities and uses it as the standard entity associated with the first entity. The aforementioned multi-factor fusion model is an existing model. This model extracts discriminative factor features from the knowledge base, initial element pool, and candidate entities, and then scores each candidate entity based on these discriminative factor features, using the candidate entity with the highest score as the standard entity associated with the first entity. The discriminative factor features mentioned above may include textual semantic similarity features (representing the textual semantic similarity between the context text of the first entity and the descriptive text of the candidate entity), geographic location features (representing the distance between the geographic location of the first entity and the geographic location of the candidate entity), type consistency features (representing whether the type of the first entity and the type of the candidate entity are consistent), and co-existing entity association features (representing whether the candidate entity has a knowledge graph relationship with other linked entities in the element pool), etc. The discriminative factor features mentioned above affect the score of the candidate entity. For example, the larger the feature value corresponding to the textual semantic similarity feature of the first entity and the candidate entity, the more similar the first entity and the candidate entity are, and the higher the score of the candidate entity.

[0195] The aforementioned conflict detection is used to verify the rationality of relationships between entities, in order to identify and eliminate redundant and erroneous information in knowledge fragments. These relationships can refer to the association between the extracted entities and standard entities, or they can refer to the relationships between the extracted entities themselves. These relationships can be semantic relationships, such as those related to "located in," "belonging to," or "associated with," or they can be semantically consistent. Specifically, semantic relationships between entities can be extracted using pre-trained models or pre-set algorithms. When performing conflict detection on the association between extracted entities and standard entities, this can be done by checking whether the content of a specific type of entity is consistent with the content of the standard entity in the knowledge base. For example, if the founding date of Block A is 1887 (time entity: 1887), but the founding date of Block A in the knowledge base is 1905 (standard entity: 1905), checking whether the time entity and the standard entity are consistent reveals an inconsistency and a conflict. In this case, the content of the extracted entities can be modified based on the knowledge data in the knowledge base. The process of conflict detection between extracted entities is similar to the process of conflict detection between extracted entities and standard entities, and will not be repeated here. Furthermore, when the semantic relationships between extracted entities are semantically consistent, entities with such semantic relationships (i.e., semantically consistent) can be merged to avoid element fragmentation and redundancy.

[0196] The aforementioned knowledge completion is used to add entities or relationships that failed to link to the standard entities as new knowledge to the knowledge base, thereby completing the knowledge base. Of course, this knowledge completion can also be used to complete the street elements in the element pool based on multi-source knowledge data from various knowledge bases. Specifically, it can complete street elements for element types that do not exist in the element pool, or it can complete street elements already in the element pool. Specifically, it can determine the element types corresponding to the street elements included in the element pool, and from the preset standard element types, determine the element types that do not exist in the element pool and use them as element types to be supplemented. From the multi-source knowledge data (i.e., from each knowledge base), determine the elements corresponding to the element types to be supplemented and add them to the element pool as street elements. Alternatively, when the content between the aforementioned first entity and the standard entities that are related to the first entity is inconsistent, the content of the standard entities that are related to the first entity shall prevail, and the content of the first entity shall be completed or modified accordingly.

[0197] The aforementioned knowledge verification model can be a large language model, such as the Qwen3-32B model, which also has hundreds of billions of parameters. Alternatively, it can be a large model pre-trained by fine-tuning a general large model. During training, a first training sample and a first sample label are obtained. The first training sample is the element pool, and the first sample label is the element pool after manual verification of the street elements in the element pool. The first training sample is input into the general large model to obtain the first element pool after verification by the general large model. Based on the first element pool and the first sample label, a pre-set verification loss function is used to determine the verification loss. The general large model is then trained based on the verification loss to obtain the knowledge verification model. This knowledge verification model is used to verify the street elements in the cleaned element pool, mainly performing contradiction detection and logical consistency verification, further eliminating duplicate or conflicting elements in the element pool to output a standardized set of street elements, i.e., the target element pool. This knowledge verification model can verify the street elements in each initial element pool after cleaning to obtain the target element pool corresponding to each street.

[0198] In some embodiments of this specification, in order to quickly acquire data of various modalities in S2 above, a distributed data acquisition method can be adopted to acquire data of various modalities from various types of data sources. The aforementioned distributed data acquisition refers to the synchronous acquisition of data of various modalities from various types of data sources by multiple distributed servers or acquisition nodes. Specifically, each distributed server or acquisition node can acquire data from each type of data source, that is, there is a one-to-one correspondence between the distributed server or acquisition node and the data source. Of course, multiple distributed servers or acquisition nodes can also acquire data from one type of data source, that is, there is a one-to-many or many-to-one relationship between the distributed server or acquisition node and the data source.

[0199] In some embodiments of this specification, after obtaining the protection effectiveness index of each block object, the blocks can be sorted in descending order of protection effectiveness index to obtain a block sequence, which is then displayed to the user. Additionally, the server can identify block objects with protection effectiveness indices greater than a preset threshold and designate them as first block objects. These first block objects are then sorted in descending order of protection effectiveness index to obtain a first block sequence. A first block list is generated based on the first block sequence. Simultaneously, block objects with protection effectiveness indices not greater than a preset threshold are identified as second block objects. These second block objects are then sorted in descending order of protection effectiveness index to obtain a second block sequence. A second block list is generated based on the second block sequence. The first and second block lists are then displayed to the user. Of course, the server can also predict the future protection effectiveness of each block object based on its protection effectiveness index, and can also upload the protection effectiveness index of each block object to the historical and cultural block governance platform to provide data support for the platform, thereby better governing historical and cultural blocks.

[0200] Furthermore, after obtaining the protection effectiveness index for each block, the maintenance plan for each block can be optimized based on this index. Specifically, the protection effectiveness index for each block can be sent to relevant management personnel, who can then optimize the maintenance plan for each block based on this index. In particular, optimization can be performed on the maintenance plans for blocks with low protection effectiveness indices to better protect these blocks.

[0201] In some embodiments of this specification, the data in the various types of data sources mentioned above are updated in real time. New dimensions of data may appear in the various types of data sources mentioned above. Therefore, the original multimodal data or target multimodal data that needs to be re-acquired in S2 above may have changed in dimensions. That is, new dimensions of data appear in the original multimodal data or target multimodal data. Therefore, the server can redetermine each original evaluation index based on each dimension of the new target multimodal data, and redetermine each target evaluation index based on each original evaluation index using principal component analysis. The server can then reconstruct a multi-level indicator system for evaluating the effectiveness of street protection to update the multi-level indicator system.

[0202] The above description is merely an embodiment of this specification and is not intended to limit this specification. Various modifications and variations can be made to this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of the claims of this specification.

Claims

1. A method for evaluating the effectiveness of historical and cultural block preservation based on big data, characterized in that, The method includes: S1: Generate street elements based on a large model: Obtain a list of street names and task prompts and input them into a large model for generating street elements. Enhance the large model for generating street elements by searching through pre-built knowledge bases. Generate street elements for each street object in the list of street names through the large model for generating street elements. Based on the street name, entity elements, spatial elements and social element categories, combine them into a target element pool as the output of the large model for generating street elements. S2: Obtain street data from multimodal data: Obtain target multimodal data from the data source based on different data processing strategies; Collect data matching the street elements in the target element pool of each street object from the target multimodal data, and use it as street data; S3: Construct a multi-level indicator system for evaluating the effectiveness of neighborhood protection: From the various dimensions of the target multimodal data, determine the dimensions used to evaluate the effectiveness of neighborhood protection as each original evaluation indicator, and based on the original evaluation indicators, use principal component analysis to determine each target evaluation indicator, and construct a multi-level indicator system for evaluating the effectiveness of neighborhood protection. S4: Evaluate the effectiveness of the protection of historical and cultural blocks: Based on the block data and the multi-level indicator system for evaluating the effectiveness of block protection, evaluate the effectiveness of the protection of each block.

2. The method for evaluating the effectiveness of historical and cultural block protection based on big data as described in claim 1, characterized in that, The knowledge base includes a geographic information database, a neighborhood encyclopedia, a historical document database, and a conservation planning knowledge base.

3. The method for evaluating the effectiveness of historical and cultural block protection based on big data as described in claim 1, characterized in that, S1 specifically includes: Obtain a list of neighborhood names and task prompts; The block name list and the task prompt words are input into the block element generation model. The block element generation model is enhanced by searching through pre-built knowledge bases. The block element generation model generates block elements for each block object in the block name list. The block elements of each block object are divided based on block name, entity elements, spatial elements and social element categories. The results of the division of each block object are combined into the initial element pool of each block object and used as the output of the block element generation model. Perform feature cleaning on the block features in the initial feature pool of each block object; The knowledge verification model is used to verify the street elements in each initial element pool after cleaning, so as to obtain the target element pool for each street object.

4. The method for evaluating the effectiveness of historical and cultural block protection based on big data as described in claim 1, characterized in that, The multi-level indicator system for assessing the effectiveness of neighborhood protection includes primary assessment indicators, secondary assessment indicators, and tertiary assessment indicators. Each primary assessment indicator includes a first number of secondary assessment indicators, and each secondary assessment indicator includes a second number of tertiary assessment indicators.

5. The method for evaluating the effectiveness of historical and cultural block protection based on big data as described in claim 4, characterized in that, S4 specifically includes: S41: Based on the street data of each street object, determine the index value of each third-level evaluation index corresponding to each street object; S42: Construct the original matrix of the three-level evaluation indicators based on the indicator values ​​of each of the three-level evaluation indicators for each block object; S43: Based on the original matrix of the three-level evaluation indicators, determine the target indicator weight corresponding to each evaluation indicator in the multi-level indicator system for evaluating the effectiveness of street protection; S44: Determine the protection effectiveness index corresponding to each block object based on the target indicator weights corresponding to each evaluation indicator and the original matrix of the three-level evaluation indicators.

6. The method for evaluating the effectiveness of historical and cultural block protection based on big data as described in claim 5, characterized in that, Specifically, S43 includes: S431: The three-level evaluation indicators are used as evaluation indicators to be processed, and the original matrix of the three-level evaluation indicators is used as the original matrix of evaluation indicators to be processed. S432: Regularize the original matrix of the evaluation indicators to be processed to obtain the standardized matrix of the evaluation indicators to be processed. S433: Based on the standardized matrix of the evaluation indicators to be processed, the entropy value corresponding to each evaluation indicator to be processed is calculated using the entropy method. S434: Determine the difference coefficient corresponding to each evaluation indicator to be processed based on the entropy value corresponding to each evaluation indicator to be processed; S435: Normalize the difference coefficients of each evaluation indicator to be processed under the same superior evaluation indicator to obtain the target indicator weights of each evaluation indicator to be processed under the same superior evaluation indicator. S436: When the evaluation indicator to be processed is not the first-level evaluation indicator, determine the indicator value corresponding to the superior evaluation indicator of each evaluation indicator to be processed for each block object according to the target indicator weight corresponding to each evaluation indicator to be processed and the standardized matrix of the evaluation indicator to be processed, and construct the original matrix of the superior evaluation indicator. S437: The superior evaluation indicator is used as the evaluation indicator to be processed again, and the original matrix of the superior evaluation indicator is used as the original matrix of the evaluation indicator to be processed again. Steps S432 to S437 are repeated until the evaluation indicator to be processed is a first-level evaluation indicator. The target indicator weights corresponding to each evaluation indicator in the multi-level indicator system for evaluating the effectiveness of the protection of the street are obtained.

7. The method for evaluating the effectiveness of historical and cultural block protection based on big data as described in claim 6, characterized in that, Specifically, S432 includes: When the original matrix of the evaluation indicators to be processed is the original matrix of the third-level evaluation indicators, a preset first regularization formula is used to regularize the indicator values ​​corresponding to the positive evaluation indicators in the original matrix of the evaluation indicators to be processed, and a preset second regularization formula is used to regularize the indicator values ​​corresponding to the negative evaluation indicators in the original matrix of the evaluation indicators to be processed, so as to obtain the standardized matrix of the evaluation indicators to be processed. When the original matrix of the evaluation indicators to be processed is not the original matrix of the third-level evaluation indicators, the first regularization formula is used to regularize the indicator values ​​corresponding to all evaluation indicators in the original matrix of the evaluation indicators to be processed.

8. The method for evaluating the effectiveness of historical and cultural block protection based on big data as described in claim 6, characterized in that, Specifically, S435 includes: When the evaluation index to be processed is the secondary evaluation index, the difference coefficients corresponding to each evaluation index to be processed under the same superior evaluation index are normalized to determine the initial index weights corresponding to each evaluation index to be processed under the same superior evaluation index. Based on the Delphi method, the initial indicator weights corresponding to the first evaluation indicator among the evaluation indicators to be processed are adjusted to obtain the target indicator weights corresponding to the first evaluation indicator; the first evaluation indicator is determined according to the importance level and data fluctuation of each evaluation indicator to be processed. Identify the assessment indicators to be processed that belong to the same superior assessment indicator as the first assessment indicator, and use them as related assessment indicators. Based on the target indicator weight corresponding to the first evaluation indicator, the initial indicator weight corresponding to the associated evaluation indicator is adjusted to obtain the target indicator weight corresponding to the associated evaluation indicator. The initial indicator weights corresponding to the other evaluation indicators to be processed, excluding the first evaluation indicator and the associated evaluation indicator, are used as the target indicator weights.

9. The method for evaluating the effectiveness of historical and cultural block protection based on big data as described in claim 5, characterized in that, In step S44, determining the protection effectiveness index for each street block object based on the target indicator weights corresponding to each evaluation indicator and the original matrix of the three-level evaluation indicators specifically includes: The original matrix of the three-level evaluation indicators is regularized to obtain the standardized matrix of the three-level evaluation indicators. Based on the standardized matrix of the three-level evaluation indicators, the third indicator value corresponding to each of the three-level evaluation indicators for each block object is determined; Using the target indicator weight of each of the three-level evaluation indicators, the third indicator value of each of the three-level evaluation indicators of each block object is weighted to obtain the third weighted value of each of the three-level evaluation indicators of each block object; Based on the third weighted value of each of the three-level evaluation indicators for each block object, determine the second indicator value of each of the two-level evaluation indicators for each block object; Using the target indicator weight of each secondary evaluation indicator, the second indicator value of each secondary evaluation indicator of each block object is weighted to obtain the second weighted value of each secondary evaluation indicator of each block object; Based on the second weighted value of each secondary evaluation indicator for each block object, determine the first indicator value of each primary evaluation indicator for each block object; Using the target indicator weight of each primary evaluation indicator, the first indicator value of each primary evaluation indicator of each block object is weighted to obtain the first weighted value of each primary evaluation indicator of each block object; The sum of the first weighted values ​​of each primary evaluation indicator for each block object is determined and used as the corresponding protection effectiveness index for each block object.

10. The method for evaluating the effectiveness of historical and cultural block protection based on big data as described in claim 1, characterized in that, The specific steps in S2, which involve obtaining target multimodal data from the data source based on different data processing strategies, include: We acquire data of various modalities from various types of data sources and use it as raw multimodal data. Based on different data processing strategies, the multimodal data is processed to obtain target multimodal data; the data processing strategies include data cleaning, data semantic processing, media rating classification, sentiment analysis, and event extraction and recognition.

11. The method for evaluating the effectiveness of historical and cultural block protection based on big data as described in claim 1, characterized in that, In step S3, based on the original evaluation indicators, principal component analysis is used to determine the target evaluation indicators and construct a multi-level indicator system for evaluating the effectiveness of street protection. Specifically, this includes: Based on the target multimodal data, determine the original indicator value corresponding to each original evaluation indicator of each block object; Determine the original data matrix consisting of the original index values ​​corresponding to each original evaluation index of each block object; Each data element in the original data matrix is ​​standardized to obtain the target data matrix; Based on the target data matrix, the correlation coefficient matrix is ​​calculated using principal component analysis, and the correlation coefficient matrix is ​​decomposed to obtain the eigenvalues ​​and eigenvectors corresponding to each principal component. Based on the eigenvalues ​​of each principal component, the principal components are screened to obtain the target principal components; Identify the elements in the eigenvector corresponding to each target principal component whose loading reaches a specified threshold, and use them as target elements; Determine the original evaluation indicators corresponding to each target element, and use them as the evaluation indicators for each target; Based on the aforementioned target evaluation indicators, a multi-level indicator system for evaluating the effectiveness of neighborhood protection is constructed.

Images