Unified organization, expression and retrieval method of multi-heterogeneous spatiotemporal elements of silk heritage

Abstract

The present invention discloses a unified organization, expression and retrieval method for multi-heterogeneous spatiotemporal elements of silk heritage, comprising the following steps: firstly, analyzing the temporal, spatial and thematic attribute characteristics of silk heritage elements, and establishing an extensible metadata model; then realizing polymorphic storage and unified organization of metadata and unstructured data; then, based on the spatiotemporal and thematic characteristics, constructing a temporal and spatial expression model of silk heritage elements, as well as spatiotemporal and thematic associations; finally, realizing spatiotemporal retrieval of silk heritage elements, thematic retrieval and similar object recommendation. The present invention has broad application prospects and practical value in the organization and management of cultural heritage spatiotemporal big data and multi-scale spatiotemporal expression and retrieval.

Description

Technical Field

[0001] The present invention relates to the field of data organization and retrieval, and in particular to a method for uniformly organizing, expressing and retrieving multi-dimensional and heterogeneous spatiotemporal elements of silk heritage. Background Art

[0002] Silk is one of the important representatives of the long history of Chinese culture. The Silk Road, with silk as its driving force, has greatly promoted the exchange and integration of Eastern and Western culture, economy and other aspects, and has made important contributions to the development of human civilization. With the Silk Road as its framework, the global silk heritage is rich and diverse, and the diversity of silk heritage is determined by mulberry planting, silkworm breeding, silk reeling, weaving, printing and dyeing to sewing silk products. The multiple categories of silk heritage are created, including archaeological relics, documentary records, image remains, production tools, silk dyes, etc.; the silk heritage has a wide distribution from the Neolithic Age to the present, and is spread across all continents in the world.

[0003] On the one hand, the current information construction of silk heritage is facing the problems of large data volume, wide sources, and a wide variety of topics. Due to the coexistence of structured, semi-structured and unstructured data and the multi-heterogeneous properties of silk heritage data, the traditional cultural heritage data organization and expression cannot adapt to the storage and unified organization of data. On the other hand, the spatiotemporal distribution of silk heritage data is extensive and the spatiotemporal scale is inconsistent. It contains rich spatiotemporal information and rich image patterns, dye components and thematic attribute information. In-depth analysis of the correlation between the spatiotemporal data of silk heritage and the establishment of a multi-dimensional similarity calculation method are effective models for exploring the correlation between silk heritage elements.

[0004] Based on this, a polymorphic storage and unified organization method for multiple silk heritage data is constructed from multiple dimensions to achieve multi-scale retrieval of silk heritage data and integrated management, expression and retrieval of multiple heterogeneous silk heritage data results. Summary of the invention

[0005] The purpose of the present invention is to overcome the problems of existing silk heritage data organization, expression and retrieval, and to provide a unified organization, expression and retrieval method for multiple heterogeneous silk heritage spatiotemporal elements.

[0006] In order to achieve the above-mentioned invention object, the technical solution specifically adopted by the present invention is as follows:

[0007] A method for uniformly organizing, expressing and retrieving multiple heterogeneous spatiotemporal elements of silk heritage, comprising the following steps:

[0008] S1: Establish an extensible metadata entity model based on the temporal, spatial and thematic attributes of silk heritage elements;

[0009] S2: Data storage of multiple heterogeneous silk heritage elements, wherein metadata is stored in a relational database according to the metadata entity model, and unstructured data is stored in a distributed object storage system in an associative manner, thereby achieving polymorphic storage and unified organization;

[0010] S3: For the multi-heterogeneous silk heritage elements after storage, a temporal and spatial expression model of silk heritage elements is constructed based on temporal, spatial and thematic characteristics to display the temporal and spatial distribution of silk elements and the temporal and spatial change laws;

[0011] S4, traverse the multi-heterogeneous silk heritage elements after storage, calculate the temporal correlation, spatial correlation and thematic correlation of any two silk heritage elements, and obtain the comprehensive correlation through weighted summation;

[0012] S5: When a search request is received, the search conditions are parsed from the search request, and a spatiotemporal search or a thematic search of the silk heritage elements is performed according to the type of the search conditions, and the search results are returned; at the same time, when a similar object recommendation request is received, the comprehensive correlations between the current silk heritage element and other silk heritage elements are traversed and sorted, and similar objects are recommended in descending order of comprehensive correlations.

[0013] Preferably, the specific implementation method of S1 is:

[0014] S11: For all the multi-heterogeneous silk heritage elements, according to the preset thematic classification, extract the common attributes of the silk heritage elements of different themes, and build a basic metadata entity model, where the common attributes include the identification information, content information, distribution information and other metadata information of each silk heritage element;

[0015] S12: For each thematic feature, customize and extend the thematic design attribute description to form the thematic metadata specification.

[0016] Preferably, in the metadata entity model, the common attribute unique identifier serial number is used as the data reference identifier, the geographical location information corresponding to the discovery place, collection place or origin of the silk heritage element is used as the geographical description identification information, the silk heritage element name, age, keywords, instructions and references are used as content information fields, the unstructured data online path and original path are used as distribution metadata information, and the data creation date, update date, contact person and contact information are used as other metadata information.

[0017] Preferably, the specific implementation method of S2 is:

[0018] S21: Clean, convert and normalize the metadata of the multi-heterogeneous silk heritage elements, and then store them in the spatial database PostgreSQL by topic;

[0019] S22: The unstructured data in the multi-dimensional and heterogeneous silk heritage elements are divided into a multi-level data resource directory according to thematic type, era and region, and stored in a distributed object storage system;

[0020] S23: Through the unique identification code of the silk heritage elements, the metadata stored in the spatial database PostgreSQL and the unstructured data stored in the distributed object storage system are associated and organized.

[0021] Preferably, the specific implementation method of S3 is:

[0022] S31. Convert the age fields of all silk heritage elements into AD years through the time semantic conversion model, and use this as a unified time semantic expression benchmark to construct a time axis that includes the complete age distribution range of silk heritage elements. Map the AD age interval endpoints corresponding to each silk heritage element onto the time axis to display the time distribution and association between elements.

[0023] S32. Divide the world into regions according to preset rules, and obtain historical maps and place name change relationships from different historical periods; obtain the geographical description and identification information of each silk heritage element, establish point elements on the map according to the geographical location information of the discovery, collection or production place recorded therein, and provide multi-scale spatial expressions of points, points combined with surfaces, or points, lines and surfaces according to the set display requirements, to display the point distribution, regional distribution and temporal and spatial change patterns of silk elements.

[0024] Preferably, in the multi-scale spatial expression, for places that can be accurately searched, they are directly expressed in the form of points on the map; for places of origin that cannot be accurately described, if the place of origin is at the national level or above, a center point is generated according to the regional vector range of the place of origin, and it is expressed in the form of a combination of points and surfaces and diffusion with the point as the center; if the place of origin is at the national level or below, it is still expressed in the form of points; the spatiotemporal laws of the origin, exchange and dissemination of the silk heritage are expressed in the form of a combination of points, lines and surfaces.

[0025] Preferably, the specific implementation method of S4 is:

[0026] S41: For the multivariate heterogeneous silk heritage elements after storage, calculate the temporal correlation between any two silk heritage elements

[0027]

[0028] Among them, α nis the time-related attenuation factor, ranging from 0 to 1; [t1, t2] is the age range of the i-th silk heritage element, and [t3, t4] is the age range of the j-th silk heritage element;

[0029] S42: For the multivariate heterogeneous silk heritage elements after storage, calculate the spatial correlation between any two silk heritage elements

[0030]

[0031] Among them, β is the spatial correlation attenuation factor, ranging from 0 to 1; J (Spatial a ,Spatial b ) and B(Spatial a ,Spatial b ) represent the spatial extent of the two silk heritage elements. a and Spatial b The intersection and union between ;

[0032] S43. For the multivariate heterogeneous silk heritage elements after storage, the silk pattern similarity, silk dye similarity and thematic attribute similarity of any two silk heritage elements are calculated, wherein the silk pattern similarity is obtained by extracting features of silk patterns through a convolutional neural network and calculating the similarity between feature graphs, the silk dye similarity is obtained by encoding the silk dye composition and color information through word vectors and calculating the similarity between word vectors, and the thematic attribute similarity is obtained by encoding the silk thematic attribute field information through word vectors and calculating the similarity between word vectors; and the weighted sum of the silk pattern similarity, silk dye similarity and thematic attribute similarity is taken as the thematic association P between the two silk heritage elements. sim :

[0033] P sim =ω 1 ×P pattern +ω 2 ×P dye +ω 3 ×P attribute

[0034] Among them, ω 1 ,ω 2 ,ω 3 Respectively represent the weights of silk pattern similarity, silk dye similarity and topic attribute similarity, ω 1 +ω 2 +ω 3 =1;

[0035] S44, traverse the multivariate heterogeneous silk heritage elements after storage, and construct a comprehensive correlation matrix that records the comprehensive correlation between any two silk heritage elements, where the comprehensive correlation between two silk heritage elements P is the corresponding time correlation P A , spatial correlation P S Relevance to topic P sim The weighted sum of:

[0036] P=ω A ×P A +ω S ×P S +ω sim ×P sim

[0037] Among them, ω A ,ω S ,ω sim Respectively represent the weights of silk pattern similarity, silk dye similarity and topic attribute similarity, ω A +ω S +ω sim =1.

[0038] Preferably, the specific implementation method of S5 is:

[0039] S51. The server receives the search request sent by the client in real time. When receiving the search request, the server parses the search condition and type from the search request, and returns the corresponding search result according to the search condition and type, wherein:

[0040] If the search condition type is time search, all silk heritage elements whose ages match the time search condition T in the search request are retrieved from the database and returned to the client as search results;

[0041] If the search condition type is spatial search, all silk heritage elements whose geographical description identification information matches the spatial search condition S in the search request are retrieved from the database and returned to the client as the search result;

[0042] If the search condition type is spatiotemporal search, first parse the time search condition T and the space search condition S from the spatiotemporal search request, then retrieve all silk heritage elements whose ages match the time search condition T from the database as intermediate results, and finally retrieve all silk heritage elements whose geographic description identification information matches the space search condition S from the intermediate results and return them to the client as search results;

[0043] S52. The server receives in real time a similarity recommendation request sent by the client. When receiving the similarity recommendation request, the server parses the target silk heritage element to be recommended from the similarity recommendation request, searches for the comprehensive correlation between the target silk heritage element and each other silk heritage element from the comprehensive correlation matrix, sorts all the comprehensive correlations found, and selects several silk heritage elements with the highest ranking in descending order of comprehensive correlation as recommendation results and returns them to the client.

[0044] Preferably, when calculating the similarity of the silk patterns, all silk patterns are first subjected to uniform normalization, denoising, and segmentation preprocessing to form an image dataset C with consistent resolution, and then a convolutional neural network is used to extract image features from the image dataset C, and principal component analysis is used to perform feature compression to obtain a feature map of each silk pattern, and finally the Minkowski distance is used to measure the silk pattern similarity between the feature maps.

[0045] Preferably, when calculating the silk dye similarity and the topic attribute similarity, the word2vec model is used for word vector encoding.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention expands the storage method and associated expression of the organization of traditional cultural heritage data, and effectively realizes the unified organization storage and multi-dimensional space-time and attribute association of multi-dimensional heterogeneous silk heritage data. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flowchart of the steps of the unified organization expression and retrieval method of the multi-heterogeneous silk heritage spatiotemporal elements of the present invention;

[0049] Figure 2 It is a schematic diagram of the organization of the unified resource directory of unstructured data of the present invention;

[0050] Figure 3 It is a schematic diagram of the temporal semantic conversion model of the present invention;

[0051] Figure 4 It is a time correlation schematic diagram of the present invention;

[0052] Figure 5 It is a spatial association schematic diagram of the present invention;

[0053] Figure 6 It is a schematic diagram of the subject association calculation process of the present invention; DETAILED DESCRIPTION

[0054] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] like Figure 1 As shown, in a preferred embodiment of the present invention, a method for uniformly organizing, expressing and retrieving multi-heterogeneous spatiotemporal elements of silk heritage is provided, which comprises the following steps:

[0056] S1: Establish an extensible metadata entity model based on the temporal, spatial and thematic attribute characteristics of silk heritage elements.

[0057] In the present invention, the specific implementation method of S1 is:

[0058] S11: For all the multi-heterogeneous silk heritage elements, according to the preset subject classification, extract the common attributes of the silk heritage elements of different subjects and construct a basic metadata entity model, where the common attributes include the identification information, content information, distribution information and other metadata information of each silk heritage element.

[0059] As an example of the present invention, the silk heritage data topic can be divided into seven categories: silk cultural relics, silk images, silk historical documents, silk crafts, sites, dyes and fibers.

[0060] As an example of the present invention, in the metadata entity model, the common attribute unique identifier serial number is used as the data reference identifier, the geographical location information of the silk heritage element (such as the unearthed / discovered place, collection place and origin) is used as the geographical description identification information, the silk heritage element name, age, keywords, description and references are used as content information fields, the unstructured data online path and original path are used as distribution metadata information, and the data creation date, update date, contact person and contact information are used as other metadata information.

[0061] S12: For each thematic feature, customize and extend the thematic design attribute description to form the thematic metadata specification.

[0062] As an example of the present invention, due to the particularity of thematic design attributes, extensible metadata information is designed according to thematic features. It should be noted that for each topic, the custom extended thematic design attributes can be adjusted according to actual conditions. Taking silk cultural relics as an example, the material, technology, pattern theme and size of the cultural relics are used as thematic design attributes to form the metadata specifications of the seven major topics.

[0063] S2: The data of multiple heterogeneous silk heritage elements are stored in the database, wherein the metadata is stored in the relational database according to the metadata entity model, and the unstructured data is stored in the distributed object storage system in an associated manner, thereby realizing polymorphic storage and unified organization.

[0064] In the present invention, the specific implementation method of S2 is:

[0065] S21: Clean, convert and normalize the metadata of the multi-heterogeneous silk heritage elements, and then store them in the spatial database PostgreSQL by topic;

[0066] S22: The unstructured data in the multi-dimensional and heterogeneous silk heritage elements are divided into a multi-level data resource directory according to thematic type, era and region, and stored in a distributed object storage system;

[0067] S23: Through the unique identification code of the silk heritage elements, the metadata stored in the spatial database PostgreSQL and the unstructured data stored in the distributed object storage system are associated and organized.

[0068] As an example of the present invention, each topic of silk heritage is entered into the PostgreSQL database with a serial number as a unique identifier after passing the data quality standardization check, and the topic data is stored in the corresponding metadata table; the unstructured data is mainly pictures and videos, and the unstructured data is named with a unique identifier_serial number, and is uniformly stored in a distributed object storage system in a multi-level hierarchical directory manner according to the topic category, era and region; thereby, the association organization and access of metadata and unstructured data are realized based on the unique identifier.

[0069] S3: For the multi-heterogeneous silk heritage elements after storage, a temporal and spatial expression model of the silk heritage elements is constructed based on the temporal, spatial and thematic characteristics to display the temporal and spatial distribution and temporal and spatial variation patterns of the silk elements.

[0070] In the present invention, the specific implementation method of S3 is:

[0071] S31. Convert the age fields of all silk heritage elements into AD years through the time semantic conversion model, and use this as a unified time semantic expression benchmark to construct a time axis that includes the complete age distribution range of silk heritage elements. Map the AD age interval endpoints corresponding to each silk heritage element onto the time axis to display the time distribution and association between elements.

[0072] S32. Divide the world into regions according to preset rules, and obtain historical maps and place name change relationships from different historical periods; obtain the geographical description and identification information of each silk heritage element, establish point elements on the map according to the geographical location information of the discovery, collection or production place recorded therein, and provide multi-scale spatial expressions of points, points combined with surfaces, or points, lines and surfaces according to the set display requirements, to display the point distribution, regional distribution and temporal and spatial change patterns of silk elements.

[0073] It should be noted that the standard time form converted by the time semantic conversion model can be adjusted according to actual conditions, and the length range of the time axis can also be adjusted according to the time period required to be displayed. As an example of the present invention, for time expression, the division of the years is based on the order of Chinese historical dynasties, and the time span from prehistoric 5000 to the 21st century is selected according to the spatiotemporal distribution characteristics of silk heritage data. Multivariate and heterogeneous descriptions of the years, such as the reign titles at home and abroad, the stem-branch calendar, and the Gregorian calendar are unified through time semantic conversion to use the Gregorian calendar as a unified time semantic expression benchmark.

[0074]

[0075] Furthermore, the age field of all silk heritage elements cannot be empty. The age of all elements is mapped to the timeline through semantic conversion, including the start time t1 and the end time t2, to determine the association between the time intervals [t1, t2] of different elements. For example, if elements a and b correspond to the time periods [t1, t2] and [t3, t4] respectively, the time association between the two elements exists in the following states: before, after, contains, meet, overlap, and equal. The specific judgment logic is as follows:

[0076] [t1,t2]before[t3,t4],t2 <t3

[0077] [t1,t2]after[t3,t4],t4 <t1

[0078] [t1,t2]meet[t3,t4],t2=t3 or t1=t4

[0079] [t1,t2]equal[t3,t4],t1=t2 and t3=t4

[0080] [t1,t2]overlap[t3,t4],t1 <t3<t2<t4 or t3<t1<t4<t2

[0081] [t1,t2]contains[t3,t4],t1 <t3<t4<t2 or t3<t1<t2<t4

[0082] As an example of the present invention, for spatial expression, in the specifically adopted multi-scale spatial expression method, for places that can be accurately searched, they can be directly expressed in the form of points on the map; for places of origin that cannot be accurately described, if the place of origin is at the national level or above, the center point can be generated according to the regional vector range of the place of origin, and expressed in the form of diffusion with the point as the center; if the place of origin is at the national level or below, it can still be expressed in the form of points; for the spatiotemporal laws of the origin, exchange, and dissemination of the silk heritage, it can be expressed in the form of a combination of points, lines, and surfaces.

[0083] For example, the first-level regions can be divided into Central Asia, East Asia, South Asia, North Asia, West Asia, Europe, North America, South America and Africa, and the regions in different regions are divided according to countries. Collect publicly released territorial maps and ancient place names of various periods around the world to form historical maps and place name change tables divided by dynasties, which are convenient for displaying territory changes and place name searches. Furthermore, at least one field in the unearthed / discovered place, collection place and origin of all silk heritage elements cannot be empty. For precisely described location information, such as the China National Silk Museum and the Metropolitan Museum of Art, the relevant longitude and latitude are obtained to construct the point element Point(lon,lat); for large-scale areas at the intercontinental level, such as those described as Europe and Asia, the center point is generated after obtaining the approximate vector range of the area, and the information is expressed in the form of a combination of points and surfaces, with the point as the center and then diffused and migrated in sequence; the spatiotemporal laws of the origin, exchange, and dissemination of silk heritage are expressed in the form of a combination of points, lines, and surfaces. For example, for the laws describing the origin of silk, silk originated in the Yellow River basin more than 5,000 years ago, and was later discovered in the Yangtze River basin. It spread throughout the country and abroad about 2,000 years ago, and is now spread all over the world. The trajectory of its origin and gradual spread outward is presented in the form of a combination of points, lines, and surfaces.

[0084] In addition, the spatial association between elements can be judged through topological relationships. Based on the spatialization of locations such as unearthed / discovered, collected or produced, all silk heritage elements are placed on the map, and the spatial association between different elements is judged through the topological relationship of points, lines and surfaces. The spatial association between two elements mainly exists in the states of inclusion, proximity, intersection and equality.

[0085] S4. Traverse the multivariate heterogeneous silk heritage elements after storage, calculate the temporal correlation, spatial correlation and thematic correlation of any two silk heritage elements, and obtain the comprehensive correlation through weighted summation.

[0086] In the present invention, the specific implementation method of S4 is:

[0087] S41: For the multivariate heterogeneous silk heritage elements after storage, calculate the temporal correlation between any two silk heritage elements

[0088]

[0089] Among them, α n is the time-related attenuation factor, ranging from 0 to 1; [t1, t2] is the age range of the i-th silk heritage element, and [t3, t4] is the age range of the j-th silk heritage element;

[0090] S42: For the multivariate heterogeneous silk heritage elements after storage, calculate the spatial correlation between any two silk heritage elements

[0091]

[0092] Among them, β is the spatial correlation attenuation factor, ranging from 0 to 1; J (Spatial a ,Spatial b ) and B(Spatial a ,Spatial b ) represent the spatial extent of the two silk heritage elements. a and Spatial b The intersection and union between ;

[0093] Based on the temporal correlation and spatial correlation, the temporal and spatial correlation can be weighted:

[0094]

[0095] Where: γ ij is the spatiotemporal correlation between data i and j, C t , C k They are the weighted parameters of temporal and spatial correlation, ranging from 0 to 1.

[0096] S43. For the multivariate heterogeneous silk heritage elements after storage, the silk pattern similarity, silk dye similarity and thematic attribute similarity of any two silk heritage elements are calculated, wherein the silk pattern similarity is obtained by extracting features of silk patterns through a convolutional neural network and calculating the similarity between feature graphs, the silk dye similarity is obtained by encoding the silk dye composition and color information through word vectors and calculating the similarity between word vectors, and the thematic attribute similarity is obtained by encoding the silk thematic attribute field information through word vectors and calculating the similarity between word vectors; and the weighted sum of the silk pattern similarity, silk dye similarity and thematic attribute similarity is taken as the thematic association P between the two silk heritage elements. sim :

[0097] P sim =ω 1 ×Ppattern +ω 2 ×P dye +ω 3 ×P attribute

[0098] Among them, ω 1 ,ω 2 ,ω 3 Respectively represent the weights of silk pattern similarity, silk dye similarity and topic attribute similarity, ω 1 +ω 2 +ω 3 =1; the specific values ​​of these three weights can be optimized according to the actual focus, and their respective values ​​are in the range of 0 to 1.

[0099] As a preferred implementation method of the present invention, when calculating the similarity of the silk patterns, all silk patterns are first subjected to uniform normalization, denoising, and segmentation preprocessing to form an image data set C with consistent resolution, and then a convolutional neural network is used to extract image features from the image data set C, and principal component analysis is used to perform feature compression to obtain a feature map of each silk pattern, and finally the Minkowski distance is used to measure the silk pattern similarity between the feature maps.

[0100] As an example of the present invention, the convolution process of the convolutional neural network is as follows:

[0101]

[0102] in is the convolutional feature map of layer l and j channels, M J is the set of feature maps of layer l-1, is the weight matrix, for The refined convolution extraction formula is:

[0103]

[0104] Where D is the depth of the convolution, F is the width and height of the convolution, and w d,m,n is the weight of layer d with m rows and n columns. After convolution, the new image dataset C is obtained. The image dataset C is traversed. i As an example of the image to be analyzed, the Minkowski distance metric C i The similarity with the other image features in the dataset, the obtained similarity ranking matrix P pattern ; The Minn distance between two points x and y in n-dimensional space is:

[0105]

[0106] Where p is a constant.

[0107] As a preferred implementation of the present invention, when calculating the silk dye similarity and the topic attribute similarity, the word2vec model is used for word vector encoding.

[0108] As an example of the present invention, the silk dye similarity mainly judges the field information of the dye composition and dye color of the silk used in the silk heritage, and the thematic attribute similarity judgment can judge the similarity of the silk thematic attribute field information such as name, material, weaving technology used, description and pattern theme. The silk dye similarity and the thematic attribute similarity calculation are calculated using the word2vec model. After the training is completed, the word2vec model can be used to map each word to a vector, indicating the relationship between words, and the vector is the hidden layer of the neural network. In short, it is to convert a word into the expression form of the corresponding vector to allow the machine to read the data. The encoded word vector can measure the similarity by means of cosine distance, etc., and then be used as dye similarity or thematic attribute similarity.

[0109] S44, traverse the multivariate heterogeneous silk heritage elements after storage, and construct a comprehensive correlation matrix that records the comprehensive correlation between any two silk heritage elements, where the comprehensive correlation between two silk heritage elements P is the corresponding time correlation P A , spatial correlation P S Relevance to topic P sim The weighted sum of:

[0110] P=ω A ×P A +ω S ×P S +ω sim ×P sim

[0111] Among them, ω A ,ω S ,ω sim Respectively represent the weights of silk pattern similarity, silk dye similarity and topic attribute similarity, ω A +ω S +ω sim =1.

[0112] These three weights ω A ,ω S ,ω sim The specific value of can be optimized according to the actual focus, and each value is within the range of 0 to 1. It should be noted that ω A ,ω S ,ω sim It can also be partially set to 0, for example, ω A =0,ω S =0,ωsim =1, so the comprehensive relevance is actually the topic relevance. At this time, when subsequent similar object recommendations are made based on the comprehensive relevance, the recommendations are actually made based on the topic relevance.

[0113] S5: When a search request is received, the search conditions are parsed from the search request, and a spatiotemporal search or a thematic search of the silk heritage elements is performed according to the type of the search conditions, and the search results are returned; at the same time, when a similar object recommendation request is received, the comprehensive correlations between the current silk heritage element and other silk heritage elements are traversed and sorted, and similar objects are recommended in descending order of comprehensive correlations.

[0114] In the present invention, the specific implementation method of S5 is:

[0115] S51. The server receives the search request sent by the client in real time. When receiving the search request, the server parses the search condition and type from the search request, and returns the corresponding search result according to the search condition and type, wherein:

[0116] If the search condition type is time search, all silk heritage elements whose ages match the time search condition T in the search request are retrieved from the database and returned to the client as search results;

[0117] If the search condition type is spatial search, all silk heritage elements whose geographical description identification information matches the spatial search condition S in the search request are retrieved from the database and returned to the client as the search result;

[0118] If the search condition type is spatiotemporal search, first parse the time search condition T and the space search condition S from the spatiotemporal search request, then retrieve all silk heritage elements whose ages match the time search condition T from the database as intermediate results, and finally retrieve all silk heritage elements whose geographic description identification information matches the space search condition S from the intermediate results and return them to the client as search results;

[0119] S52. The server receives in real time a similarity recommendation request sent by the client. When receiving the similarity recommendation request, the server parses the target silk heritage element to be recommended from the similarity recommendation request, searches for the comprehensive correlation between the target silk heritage element and each other silk heritage element from the comprehensive correlation matrix, sorts all the comprehensive correlations found, and selects several silk heritage elements with the highest ranking in descending order of comprehensive correlation as recommendation results and returns them to the client.

[0120] In order to further demonstrate the advantages of the present invention, the unified organization, expression and retrieval method of the multi-heterogeneous silk heritage spatiotemporal elements shown in S1 to S5 above is applied to a specific example to demonstrate the technical effects that can be achieved.

[0121] Example

[0122] In this embodiment, the basic process steps of the unified organization, expression and retrieval method of the multi-heterogeneous spatiotemporal elements of the silk heritage are as shown in S1 to S5 above, which will not be fully repeated. The specific implementation at the detail level will be mainly demonstrated below.

[0123] Step 1: According to S1 above, analyze the characteristics of silk heritage elements and establish an extensible metadata model

[0124] Silk heritage data has the characteristics of multi-layered heterogeneity, long time span and wide spatial distribution due to the multi-link production and multi-category heritage. It can be divided into seven categories: silk cultural relics, silk images, silk historical materials, weaving technology, mulberry silkworm sites, natural dyes and silk fibers. Common attributes are extracted to build a basic metadata entity model, including identification information, content information, distribution information and other metadata information. Among them, the common attribute unique identifier serial number is extracted as the data reference identifier, the unearthed / discovered place, collection place and origin are geographical description identification information, the silk heritage element name, age, keywords, description, dye composition and references are the main content information, the unstructured data online path and original path are the distribution metadata information, and the data creation date, update date, contact person and contact information are other common metadata information.

[0125] Each topic has distinct attributes. Based on the basic metadata entity, an extensible topic metadata specification is designed according to the topic design attribute characteristics. Taking silk cultural relics as an example, typical representatives include handed-down items and archaeological excavations. Different cultural relics have weaving technology, pattern themes and functional material attributes.

[0126] Step 2: According to S2 above, under a unified spatiotemporal reference framework, each topic of silk heritage is entered into the PostgreSQL database with a serial number as a unique identifier after data quality standardization inspection, and the topic data is stored in the corresponding metadata table; unstructured data is mainly pictures and videos, such as Figure 2 As shown in the figure, unstructured data is named with a unique identifier_serial number and is uniformly stored in the distributed object storage system OSS in a multi-level hierarchical directory manner according to thematic categories, years, and regions; metadata and unstructured data are associated, organized, and accessed based on the unique identifier.

[0127] The third step is to select the time span from 5000 years before history to the 21st century according to the above S3, and construct a unified time axis based on the order of Chinese historical dynasties. Figure 3As shown in the figure, in view of the multi-heterogeneous characteristics of the description of the age of silk heritage elements, the Ganzhi calendar is converted into the Gregorian calendar through the natural sequence method, the emperor's reign is converted into the Gregorian calendar through the table lookup method, and other calendars are converted into the Gregorian calendar through astronomical calendars, so as to achieve the unified time semantic expression standard of the Gregorian calendar. For example, the third year of Huangchu in Wei Dynasty → 222 AD.

[0128] like Figure 4 As shown, for silk elements a and b, their age intervals are [t1, t2] and [t3, t4] respectively. Then, in terms of time state, there are several time relationships: [t1, t2] before [t3, t4], [t1, t2] meet [t3, t4], [t1, t2] overlap [t3, t4], [t1, t2] after [t3, t4], [t1, t2] equal [t3, t4] and [t1, t2] contains / contained by [t3, t4].

[0129] Clarify the global regional divisions, including Central Asia, East Asia, South Asia, North Asia, West Asia, Europe, North America, South America and Africa, and the regions are divided according to countries in different regions. Collect publicly released territorial maps and ancient place names of various periods around the world, and form historical maps and place name change tables divided by dynasties, which are convenient for displaying territory changes and place name searches.

[0130] At the same time, the spatial expression of silk elements is mainly divided into three forms: 1) The places that can be accurately searched, such as the China Silk Museum, Zhejiang Provincial Museum, and the Metropolitan Museum of Art in New York, are expressed in the form of points on the map through spatialization; 2) For places of origin that cannot be accurately described, the national level and above, such as Central Asia, East Asia, Europe, and North America, are set. After obtaining the vector range of the region, the center point is generated and expressed in the form of a combination of points and surfaces, with the point as the center for diffusion. Countries and levels below them, such as China and Zhejiang, are still expressed in the form of points; 3); The spatiotemporal laws of the origin, exchange, and dissemination of silk heritage are expressed in the form of a combination of points, lines, and surfaces. For example, for the description of the laws of the origin and generation of silk, silk originated in the Yellow River Basin more than 5,000 years ago, and was later discovered in the Yangtze River Basin. It spread throughout the country and went abroad about 2,000 years ago, and is now spread all over the world. The trajectory of gradually spreading outward from the origin of sporadic points is presented in the form of a combination of points, lines, and surfaces.

[0131] like Figure 5 As shown in the figure, on the basis of clarifying the spatial expression of silk heritage elements, the spatial correlation between different elements is judged through the topological relationship of points, lines and surfaces. The spatial ranges of elements a and b are Spatial a Spatial b, then the spatial relationship includes equality, separation, inclusion, intersection, and connection.

[0132] The fourth step is to traverse the multivariate heterogeneous silk heritage elements after storage according to the aforementioned S4, calculate the temporal correlation, spatial correlation and thematic correlation of any two silk heritage elements and obtain the comprehensive correlation through weighted summation.

[0133] The calculation methods of temporal correlation and spatial correlation are as described above and will not be repeated here.

[0134] For topic relevance, Figure 6 As shown in the figure, the theme relevance is established from three aspects: silk patterns, silk heritage dye components and thematic attributes:

[0135] 1) Silk pattern similarity: The silk heritage image dataset B is obtained from the silk heritage thematic record elements. All images are uniformly normalized, denoised, segmented and other preprocessed to form a dataset C with consistent resolution. The image features of dataset C are extracted using a convolutional neural network, and feature compression is performed using principal component analysis. The dataset C is traversed and the images are sorted into C. i As an example of the image to be analyzed, the Minkowski distance metric C i The similarity with the other image features in the dataset, the obtained similarity ranking matrix P pattern ; The convolution process for the cultural relics image is as follows:

[0136]

[0137] in is the convolutional feature map of the cultural relic image with j channels at layer l, M J is the feature map set of the cultural relics images in the l-1 layer, is the cultural relic image weight matrix, for The bias.

[0138] Refine the convolution extraction of cultural relics images:

[0139]

[0140] Where D is the depth of the convolution, F is the width and height of the convolution, and w d,m,n is the weight of the cultural relic image convolution layer d, m rows and n columns. After convolution, the convolution image features of the cultural relic image are extracted and calculated to obtain a new data set C. The data set C is traversed to C. i As an example of the image to be analyzed, the Minkowski distance metric C i The similarity with the other image features in the dataset, the obtained similarity ranking matrix P pattern ; The Minn distance between two points x and y in n-dimensional space is:

[0141]

[0142] Where p is a constant.

[0143] 2) Silk dye similarity: Analyze the composition and color of dyes used in silk heritage and construct a silk dye composition similarity index P dye ;

[0144] 3) Similarity of thematic attributes: Based on the metadata information of the spatiotemporal elements of silk heritage, including the name, material, manufacturing technology used, description and pattern theme of silk heritage, the similarity index P of thematic attributes is calculated. attribute The above two similarities both use the word2vec model, which is a shallow two-layer neural network model used to generate word vectors. It is used for training to reconstruct linguistic word texts. In simple terms, it converts a word into a corresponding vector expression to allow the machine to read the data. Taking thematic attribute similarity as an example, the field information is:

[0145] y w =(y w,1 ,y w,2 ,…,y w,n ) T

[0146] For example, the field name ("Silk Heritage Name", "Excavation Site", "Excavation Year", "Material", "Technology Used", "Pattern Theme", "Collection Place"), one of the records corresponds to the field information record ("Round Lion Pattern Brocade", "Dunhuang Mogao Grottoes", "Five Dynasties", "Silk", "Liao-style Twill Weft Brocade", "Lion", "Victoria and Albert Museum"). The encoded word vector is calculated through the cosine distance to obtain the similarity index P of the thematic attribute. attribute .

[0147] Finally, the thematic similarity values ​​between silk heritage elements are measured based on the silk pattern similarity, silk dye similarity and thematic attribute similarity. The similarity calculation formula between elements a and b is as follows:

[0148] P sim (a,b)=ω 1 ×P pattern (a,b)+ω 2 ×P dye (a,b)+ω 3 ×P attribute (a,b)

[0149] Among them, ω1+ω2+ω3=1, which respectively represent the weights of pattern similarity, dye composition similarity and attribute similarity, which are determined according to the specific factor set test.

[0150] The comprehensive correlation between two silk heritage elements P is the corresponding time correlation P A , spatial correlation P S Relevance to topic P sim The weighted sum of:

[0151] P=ω A ×P A +ω S ×P S +ω sim ×P sim

[0152] Among them, ω A ,ω S ,ω sim Respectively represent the weights of silk pattern similarity, silk dye similarity and topic attribute similarity, ω A +ω S +ω sim =1.

[0153] Step 5: According to the aforementioned step S5, the server parses, matches and returns the results of the search request and similarity recommendation request sent by the client.

[0154] In this example, a silk interactive map platform was built based on the above method. The retrieval matching process for time retrieval in this platform is as follows: for each silk heritage thematic record constituting the element set, determine whether the excavation / discovery or recording age Ai of each element Ai matches the time retrieval condition T in a time-related relationship. Intersection, equality, and inclusion can all be considered as matches. If they match, the element Ai is returned to the client as the retrieval result A. out :

[0155]

[0156] For example, select a time period on the timeline, such as "Song Dynasty", to filter out all records that match the time period of the Song Dynasty.

[0157] When performing spatial retrieval in this platform, the retrieval matching process is as follows: determine whether the spatial position Si of the element Ai matches the spatial retrieval condition S in a spatial association relationship. Intersection, equality, and inclusion can all be considered as matches. If the match is satisfied, the element Ai is returned to the client as the retrieval result A. out :

[0158]

[0159] For example, select a location on the map, such as "Mogao Grottoes in Dunhuang, Gansu", and filter out all records that match the Mogao Grottoes in Dunhuang among the places of excavation / discovery, collection, and production.

[0160] The search and matching process for spatiotemporal search in this platform is as follows: for the spatiotemporal search condition ST, the time search condition T and the spatial search condition S need to be parsed from it, and the order of matching time first and space second is used to determine whether the excavation / discovery or recording age and spatial location of the element Ai simultaneously meet the matching relationship of intersection, equality or inclusion with the condition ST. If the match is satisfied, the element Ai is returned as the search result A. out :

[0161]

[0162] For example, you can select a time period and then search for locations, such as searching for all silk elements unearthed or produced in Asia during the Song Dynasty.

[0163] The process of recommending similar objects in the platform is as follows: when a similar object recommendation request is received, the target silk heritage element to be recommended is parsed from the similar object recommendation request, and the comprehensive correlation between the target silk heritage element and each other silk heritage element is found from the comprehensive correlation matrix, all the comprehensive correlations found are sorted, and several silk heritage elements with the highest ranking are selected in descending order of comprehensive correlation as recommendation results and returned to the client.

[0164] In addition, thematic search can also be provided here. For silk heritage element set A, the similarity between each element and the subject condition to be searched is determined through full-text search, and set A1 is returned after sorting from high to low. The obtained subject similarity recommends similar objects for each element in set A1. For example, when searching for the key subject "warp skin", all records matching warp skin are returned, and they are sorted from high to low according to the similarity. When viewing the detailed information of a certain record such as "Tangled Peony Eight Auspicious Two-color Satin Warp Skin", the top five similar objects with the highest matching degree are returned according to the subject similarity measurement rules.

[0165] The above-described embodiment is only a preferred solution of the present invention, but it is not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A unified organization, expression and retrieval method for multi-heterogeneous spatiotemporal elements of silk heritage, characterized by: The steps include: S1: Establish an extensible metadata entity model based on the temporal, spatial and thematic attributes of silk heritage elements; S2: Data storage of multiple heterogeneous silk heritage elements, wherein metadata is stored in a relational database according to the metadata entity model, and unstructured data is stored in a distributed object storage system in an associative manner, thereby achieving polymorphic storage and unified organization; S3: For the multi-heterogeneous silk heritage elements after storage, a temporal and spatial expression model of silk heritage elements is constructed based on temporal, spatial and thematic characteristics to display the temporal and spatial distribution of silk elements and the temporal and spatial change laws; S4, traverse the multi-heterogeneous silk heritage elements after storage, calculate the temporal correlation, spatial correlation and thematic correlation of any two silk heritage elements, and obtain the comprehensive correlation through weighted summation; S5: When a search request is received, the search conditions are parsed from the search request, and a spatiotemporal search or a thematic search of the silk heritage elements is performed according to the type of the search conditions, and the search results are returned; at the same time, when a similar object recommendation request is received, the comprehensive correlations between the current silk heritage element and other silk heritage elements are traversed and sorted, and similar objects are recommended in descending order of comprehensive correlations; The specific implementation method of S3 is: S31. Convert the age fields of all silk heritage elements into AD years through the time semantic conversion model, and use this as a unified time semantic expression benchmark to construct a time axis that includes the complete age distribution range of silk heritage elements. Map the AD age interval endpoints corresponding to each silk heritage element onto the time axis to display the time distribution and association between elements. S32. Divide the regions on a global scale according to preset rules, and obtain historical maps and place name change relationships of different historical periods; obtain geographical description identification information of each silk heritage element, establish point elements on the map according to the geographical location information of the discovery place, collection place or production place recorded therein, and provide multi-scale spatial expression methods of point, point-surface combination or point-line-surface combination according to the set display requirements, so as to display the point distribution, regional distribution and temporal and spatial change rules of silk elements; The specific implementation method of S4 is: S41: For the multivariate heterogeneous silk heritage elements after storage, calculate the temporal correlation between any two silk heritage elements Among them, α n is the time-related attenuation factor, ranging from 0 to 1; [t1, t2] is the age range of the i-th silk heritage element, and [t3, t4] is the age range of the j-th silk heritage element; S42: For the multivariate heterogeneous silk heritage elements after storage, calculate the spatial correlation between any two silk heritage elements Among them, β is the spatial correlation attenuation factor, ranging from 0 to 1; J (Spatial a ,Spatial b ) and B(Spatial a ,Spatial b ) represent the spatial extent of the two silk heritage elements. a and Spatial b The intersection and union between ; S43. For the multivariate heterogeneous silk heritage elements after storage, the silk pattern similarity, silk dye similarity and thematic attribute similarity of any two silk heritage elements are calculated, wherein the silk pattern similarity is obtained by extracting features of silk patterns through a convolutional neural network and calculating the similarity between feature graphs, the silk dye similarity is obtained by encoding the silk dye composition and color information through word vectors and calculating the similarity between word vectors, and the thematic attribute similarity is obtained by encoding the silk thematic attribute field information through word vectors and calculating the similarity between word vectors; and the weighted sum of the silk pattern similarity, silk dye similarity and thematic attribute similarity is taken as the thematic association P between the two silk heritage elements. sim : P sim =ω1×P pattern +ω2×P dye +ω3×P attribute Among them, ω1, ω2, and ω3 represent the weights of silk pattern similarity, silk dye similarity, and topic attribute similarity, respectively, ω1+ω2+ω3=1; S44, traverse the multivariate heterogeneous silk heritage elements after storage, and construct a comprehensive correlation matrix that records the comprehensive correlation between any two silk heritage elements, where the comprehensive correlation between two silk heritage elements P is the corresponding time correlation P A , spatial correlation P S Relevance to topic P sim The weighted sum of: P=ω A ×P A +oh S ×P S +oh sim ×P sim Among them, ω A ,ω S ,ω sim Represent the weights of silk pattern similarity, silk dye similarity and topic attribute similarity, ω A +ω S +ω sim =1.

2. The unified organization expression and retrieval method of multiple heterogeneous spatiotemporal elements of silk heritage as claimed in claim 1 is characterized in that: The specific implementation method of S1 is: S11: For all the multi-heterogeneous silk heritage elements, according to the preset thematic classification, extract the common attributes of the silk heritage elements of different themes, and build a basic metadata entity model, where the common attributes include the identification information, content information, distribution information and other metadata information of each silk heritage element; S12: For each thematic feature, customize and extend the thematic design attribute description to form the thematic metadata specification.

3. The unified organization expression and retrieval method of multiple heterogeneous spatiotemporal elements of silk heritage as claimed in claim 2 is characterized by: In the metadata entity model, the common attribute unique identifier serial number is used as the data reference identifier, the geographical location information corresponding to the discovery place, collection place or production place of the silk heritage element is used as the geographical description identification information, the silk heritage element name, age, keywords, instructions and references are used as content information fields, the unstructured data online path and original path are used as distribution metadata information, and the data creation date, update date, contact person and contact information are used as other metadata information.

4. The unified organization, expression and retrieval method for the multi-dimensional and heterogeneous spatiotemporal elements of silk heritage as claimed in claim 1 is characterized in that: The specific implementation method of S2 is: S21: Clean, convert and normalize the metadata of the multi-heterogeneous silk heritage elements, and then store them in the spatial database PostgreSQL by topic; S22: The unstructured data in the multi-dimensional and heterogeneous silk heritage elements are divided into a multi-level data resource directory according to thematic type, era and region, and stored in a distributed object storage system; S23: Through the unique identification code of the silk heritage elements, the metadata stored in the spatial database PostgreSQL and the unstructured data stored in the distributed object storage system are associated and organized.

5. The unified organization expression and retrieval method of multiple heterogeneous spatiotemporal elements of silk heritage as claimed in claim 1 is characterized in that: In the multi-scale spatial expression method, for places that can be accurately searched, they are directly expressed in the form of points on the map; for places of origin that cannot be accurately described, if the place of origin is at the national level or above, a center point is generated according to the regional vector range of the place of origin, and it is expressed in the form of a combination of points and surfaces and diffusion with the point as the center; if the place of origin is at the national level or below, it is still expressed in the form of points; the spatiotemporal laws of the origin, exchange, and dissemination of the silk heritage are expressed in the form of a combination of points, lines, and surfaces.

6. The method for uniformly organizing, expressing and retrieving the spatiotemporal elements of the multi-dimensional and heterogeneous silk heritage as claimed in claim 1, characterized in that: The specific implementation method of S5 is: S51. The server receives the search request sent by the client in real time. When receiving the search request, the server parses the search condition and type from the search request, and returns the corresponding search result according to the search condition and type, wherein: If the search condition type is time search, all silk heritage elements whose ages match the time search condition T in the search request are retrieved from the database and returned to the client as search results; If the search condition type is spatial search, all silk heritage elements whose geographical description identification information matches the spatial search condition S in the search request are retrieved from the database and returned to the client as the search result; If the search condition type is spatiotemporal search, first parse the time search condition T and the space search condition S from the spatiotemporal search request, then retrieve all silk heritage elements whose ages match the time search condition T from the database as intermediate results, and finally retrieve all silk heritage elements whose geographic description identification information matches the space search condition S from the intermediate results and return them to the client as search results; S52. The server receives in real time a similarity recommendation request sent by the client. When receiving the similarity recommendation request, the server parses the target silk heritage element to be recommended from the similarity recommendation request, searches for the comprehensive correlation between the target silk heritage element and each other silk heritage element from the comprehensive correlation matrix, sorts all the comprehensive correlations found, and selects several silk heritage elements with the highest ranking in descending order of comprehensive correlation as recommendation results and returns them to the client.

7. The unified organization, expression and retrieval method for the multi-dimensional and heterogeneous spatiotemporal elements of silk heritage as claimed in claim 1 is characterized in that: When calculating the similarity of the silk patterns, all silk patterns are first subjected to uniform normalization, denoising, and segmentation preprocessing to form an image dataset C with consistent resolution. Then, a convolutional neural network is used to extract image features from the image dataset C, and principal component analysis is used to perform feature compression to obtain a feature map of each silk pattern. Finally, the Minkowski distance is used to measure the silk pattern similarity between the feature maps.

8. The method for uniformly organizing, expressing and retrieving the spatiotemporal elements of the multi-dimensional and heterogeneous silk heritage as claimed in claim 1, characterized in that: When calculating the silk dye similarity and the topic attribute similarity, the word2vec model is used for word vector encoding.

Images