WO2022217711A1 - 基于多层关联知识图谱的信息预测方法、装置、设备及介质 - Google Patents

基于多层关联知识图谱的信息预测方法、装置、设备及介质 Download PDF

Info

Publication number
WO2022217711A1
WO2022217711A1 PCT/CN2021/097105 CN2021097105W WO2022217711A1 WO 2022217711 A1 WO2022217711 A1 WO 2022217711A1 CN 2021097105 W CN2021097105 W CN 2021097105W WO 2022217711 A1 WO2022217711 A1 WO 2022217711A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
information
historical data
knowledge graph
historical
Prior art date
Application number
PCT/CN2021/097105
Other languages
English (en)
French (fr)
Inventor
胡意仪
阮晓雯
肖京
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2022217711A1 publication Critical patent/WO2022217711A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/28—Databases characterised by their database models, e.g. relational or object models
    • G06F16/284—Relational databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/22—Indexing; Data structures therefor; Storage structures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/27—Replication, distribution or synchronisation of data between databases or within a distributed database system; Distributed database system architectures therefor
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Definitions

  • the present application relates to the technical field of knowledge graphs, and belongs to the application scenarios of intelligent information prediction based on multi-layer associated knowledge graphs in smart cities, and in particular relates to an information prediction method, device and equipment based on multi-layer associated knowledge graphs.
  • the knowledge graph can be constructed based on the massive information, and the knowledge graph can be obtained and matched with the corresponding information. associated information.
  • the inventors found that the existing knowledge graph is only suitable for storing and associating information, but cannot accurately predict the future trend based on the connection between the information. Therefore, the knowledge graph in the prior art has the problem that the trend prediction of information cannot be performed.
  • the embodiments of the present application provide an information prediction method, device, device and medium based on a multi-layer associated knowledge graph, which aims to solve the problem that the knowledge graph in the prior art cannot perform trend prediction of information.
  • an embodiment of the present application provides an information prediction method based on a multi-layer associated knowledge graph, wherein the method includes:
  • the corresponding feature information is extracted from each of the data points;
  • Layer the data points according to the characteristic information of each of the data points to obtain multiple data layers
  • the index connection information that matches the data nodes in the multi-layer relational knowledge graph is obtained as a corresponding prediction result.
  • an embodiment of the present application provides an information prediction device based on a multi-layer associated knowledge graph, wherein the information prediction device based on a multi-layer associated knowledge graph includes:
  • a historical data aggregation unit used for aggregating the historical data contained in the pre-stored historical data information according to a preset aggregation rule to obtain a plurality of corresponding data points;
  • a feature information extraction unit configured to extract corresponding feature information from each of the data points according to a preset feature extraction model and the historical data corresponding to each of the data points;
  • a data layer acquiring unit configured to obtain multiple data layers by layering the data points according to the characteristic information of each of the data points
  • a knowledge graph generation unit configured to generate a multi-layer associated knowledge graph according to the connection relationship between the historical data and the feature information of the data points in each of the data layers;
  • a newly added data feature information acquisition unit configured to obtain newly added data feature information corresponding to the newly added data information according to the feature extraction model if the newly added data information input by the user is received;
  • a judging unit configured to judge whether the multi-layer related knowledge graph contains data nodes that match the newly added data information according to preset judgment conditions and the newly added data feature information
  • a prediction result obtaining unit configured to obtain the index connection information that matches the data nodes in the multi-layer associated knowledge graph if the multi-layer associated knowledge graph includes data nodes that match the newly added data information as the corresponding prediction result.
  • an embodiment of the present application further provides a computer device, wherein the computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, the processing The computer executes the computer program to implement the information prediction method based on the multi-layer relational knowledge graph described in the first aspect.
  • an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned first aspect is implemented The information prediction method based on multi-layer relational knowledge graph.
  • Embodiments of the present application provide an information prediction method, apparatus, device, and medium based on a multi-layer associated knowledge graph. Aggregate historical data information to obtain multiple data points, extract the characteristic information corresponding to each data point and stratify the data points to obtain multiple data layers, according to the connection relationship between historical data and the data points in each data layer
  • the feature information of the corresponding multi-layer relational knowledge graph is generated, and the new feature data information corresponding to the new data information is obtained. If the multi-layer relational knowledge graph contains data nodes that match the new data information, the index corresponding to the data node is obtained.
  • the connection information is used as the prediction result.
  • a multi-layer relational knowledge graph including multiple data layers and multiple data nodes is constructed based on historical data information, and prediction results corresponding to the newly added data information are obtained based on the multi-layer relational knowledge graph. Make trend forecasts.
  • FIG. 1 is a schematic flowchart of an information prediction method based on a multi-layer associated knowledge graph provided by an embodiment of the present application
  • FIG. 2 is a schematic diagram of the effect of an information prediction method based on a multi-layer associated knowledge graph provided by an embodiment of the present application
  • FIG. 3 is a schematic sub-flow diagram of the information prediction method based on a multi-layer associated knowledge graph provided by an embodiment of the present application;
  • FIG. 4 is a schematic diagram of another sub-flow of the information prediction method based on a multi-layer associated knowledge graph provided by an embodiment of the present application;
  • FIG. 5 is a schematic diagram of another sub-flow of the information prediction method based on a multi-layer associated knowledge graph provided by an embodiment of the present application;
  • FIG. 6 is a schematic diagram of another sub-flow of the information prediction method based on a multi-layer associated knowledge graph provided by an embodiment of the present application;
  • FIG. 7 is a schematic diagram of another sub-flow of the information prediction method based on a multi-layer associated knowledge graph provided by an embodiment of the present application.
  • FIG. 8 is a schematic diagram of another sub-flow of the information prediction method based on the multi-layer associated knowledge graph provided by the embodiment of the present application.
  • FIG. 9 is a schematic block diagram of an information prediction apparatus based on a multi-layer associated knowledge graph provided by an embodiment of the present application.
  • FIG. 10 is a schematic block diagram of a computer device provided by an embodiment of the present application.
  • FIG. 1 is a schematic flowchart of an information prediction method based on a multi-layer associated knowledge graph provided by an embodiment of the present application; the information prediction method based on a multi-layer associated knowledge graph is applied to a user terminal or a management server.
  • the information prediction method of the multi-layer related knowledge graph is executed by the application software installed in the user terminal or the management server.
  • the user terminal is the terminal device that intelligently predicts the new data information input by the user, such as desktop computers, Tablet computers, mobile phones, etc.
  • the management server is the server side that receives the new data information input by the user and makes intelligent prediction, such as the server built in the enterprise or government department. As shown in FIG. 1, the method includes steps S110-S170.
  • Historical data information can be composed of multiple pieces of historical data. Each piece of historical data can be text information or image information. Historical data information contains multiple events of the same type and independent of each other. Each event can correspond to multiple pieces of historical data. The same event The multiple pieces of historical data included are connected in series and point to a final conclusion, which is the connection between the historical data. An event in the historical data information can contain only text information, or only image information, or both. Text information and image information.
  • the historical data may be historical data of meteorological data or historical data of cancer recurrence detection, and the like.
  • the event can be a rain event, and the final conclusion is rain or no rain.
  • the information collected at different time points before the final conclusion of a rain event constitutes the historical data of the event.
  • the air temperature, humidity and other information recorded at regular intervals before the event constitute the text information of the event, and the satellite cloud images obtained at regular intervals before the rain event constitute the image information of the event. Connect the collected text information and image information in series, and point to the final conclusion of rain or no rain.
  • the historical data contained in the historical data information needs to be aggregated and sorted.
  • the historical data can be aggregated through aggregation rules.
  • the aggregation rule includes a keyword set, a similarity calculation formula and a similarity threshold.
  • step S110 includes sub-steps S111 , S112 , S113 , S114 and S115 .
  • the historical data can be determined whether the historical data is text information. If it is text information, the historical data that is all text information can be aggregated. If it is not text information, the historical data that is all image information can be aggregated.
  • the keyword set includes multiple keywords of the same type as the historical data information, and multiple keywords matching the historical data of the text information can be obtained according to the associated word set.
  • the corresponding historical data is a piece of disease description information
  • the keyword set corresponding to this type may include keywords: effusion, lesions, lesions, edema, high leukocyte count, etc.
  • multiple keywords matching the text content can be obtained from the keyword set, and three keywords matching each historical data can be obtained from the keyword set. If the number of keywords matched by the data is greater than three, then according to the frequency of occurrence of each keyword in the historical data, the three keywords with the highest frequency of occurrence are obtained as the three keywords matching the historical data.
  • two pieces of historical data or pieces of historical data contain the same keyword, it means that the two pieces of historical data or pieces of historical data express the same content, and two or more pieces of historical data can be aggregated to obtain one data point, then all Each obtained data point corresponds to at least one piece of historical data.
  • the historical data is not text information
  • the historical data is a historical image
  • the image size of the historical image is the same
  • the historical image can be a color image.
  • the information type of the historical data is cancer recurrence detection
  • the corresponding historical data may be a detection image of a diseased part.
  • the similarity between the two historical images can be calculated by the similarity calculation formula. Specifically, first, grayscale processing is performed on the two historical images to obtain two grayscale images, and the grayscale value is represented by a non-negative integer.
  • the value range of the corresponding gray value of the pixel is [0, 255], the gray value of 0 means the pixel is black, the gray value of 255 means the pixel is white, and the other values indicate that the gray value is The pixel is a specific grayscale between white and black.
  • the image variance value of each grayscale image can be calculated by formula (1):
  • the image variance value of any grayscale image can be expressed as in, is the variance value of the mth row of the grayscale image, Am is the grayscale average of all pixels in the mth row of the grayscale image, B is the grayscale average of all pixels of the grayscale image, and N is the grayscale image
  • the total number of rows included for example, is the variance value of the mth row of the grayscale image F, then the image variance value includes the row-by-row variance value of the grayscale image F
  • Am is the grayscale average value of all pixels in the mth row of the grayscale image F
  • B is the grayscale average value of all pixels in the grayscale image F
  • N is the total number of lines contained in the grayscale image F.
  • N can also be the total number of columns included in the grayscale image F.
  • the similarity calculation formula can be expressed by formula (2):
  • the similarity W ⁇ between the two grayscale images is calculated, where, is the variance value of the mth row of the first grayscale image, is the variance value of the mth row of the second grayscale image.
  • the similarity between the two grayscale images is greater than the similarity threshold. If it is greater, it means that the similarity between the two historical data corresponding to the two grayscale images is greater than the similarity threshold; if not greater than , it indicates that the similarity between the corresponding two historical data is not greater than the similarity threshold. According to the calculated similarity, multiple pieces of historical data whose mutual similarity is greater than the similarity threshold can be aggregated to obtain one corresponding data point, then each data point corresponds to at least one piece of historical data.
  • the two historical data are aggregated; if the obtained similarity between the historical data FA and the historical data FB is greater than the similarity threshold, and the similarity between the historical data FB and the historical data FC is also greater than the similarity threshold, then the three historical data FA , FB and FC are aggregated.
  • Corresponding feature information is extracted from each of the data points according to a preset feature extraction model and historical data corresponding to each of the data points.
  • the feature extraction model is a specific model for extracting feature information of data points, wherein the feature extraction model includes a text encoding dictionary, a feature extraction neural network and image feature extraction rules, and the feature information of data points can be in the form of feature vectors to express.
  • step S120 includes sub-steps S121 , S122 , S123 and S124 .
  • the historical data corresponding to any data point is either text information or image information, then the type of historical data corresponding to the data point can be judged, and the historical data of each data point can be processed accordingly according to the judgment result. .
  • the multiple keywords corresponding to the data point are converted according to the text encoding dictionary, Get the encoding information for this data point. Specifically, each character can be matched to a corresponding feature code in the text coding dictionary, then the characters in the multiple keywords corresponding to each data point can be converted according to the text coding dictionary, and the converted multiple The feature codes are sequentially combined to obtain the coding information of the corresponding data points.
  • the converted coding information can be represented by a feature vector of size (1, R), then the coding information is a feature vector with 1 row and R columns, and R is a preset A fixed length value, for example, R can be set to 12, then multiple feature encoding combinations corresponding to each data point are obtained to obtain encoding information. make up.
  • S123 Input the encoded information into the feature extraction neural network for calculation to obtain feature information of each of the data points.
  • the obtained encoded information of each data point can be input into the feature extraction neural network for calculation, and the feature information of each data point can be obtained.
  • the feature extraction neural network is a neural network constructed based on artificial intelligence.
  • the feature extraction neural network consists of an input layer, multiple intermediate layers and an output layer.
  • the layers, the middle layer and the output layer are all related by association formulas.
  • c 1 and c 2 are the parameter values in the association formula.
  • the number of input nodes contained in the input layer is equal to the number of vector dimensions in the encoded information, and the vector value of each dimension in the encoded information corresponds to one input node.
  • the feature information can be represented by a feature array (R, T).
  • the size of the feature array is R row and T column, and each value in the feature array belongs to the value range of [0, 1].
  • the corresponding feature information can be extracted from the multiple pieces of image information corresponding to the data point, wherein the image feature extraction rule is to extract the corresponding feature information from multiple pieces of image information
  • the image feature extraction rules include a contrast calculation formula and a dissolution ratio value.
  • step S124 includes sub-steps S1241 , S1242 , S1243 and S1244 .
  • Each historical data corresponding to a data point is image information, and the size of the image information is the same, then multiple pieces of image information corresponding to the same data point can be superimposed.
  • the position of each pixel of the multiple pieces of image information corresponds to
  • the average calculation of the pixel value of the RGB value is to calculate the average value of the RGB value of each pixel point.
  • the RGB value contains the chromaticity values corresponding to the three color channels of red, green, and blue, respectively.
  • the average value of the pixel values corresponding to the three blue channels is obtained to obtain the pixel average value of each pixel point, and the pixel average value of each pixel point in the image information contained in the same data point is combined into a superimposed image.
  • a superimposed image is acquired and one pixel is determined as the target pixel, and eight pixels in the first layer and 16 pixels in the second layer in the periphery of the target pixel can be acquired as the target pixel , obtain the RGB value of the target pixel, the GRB value includes the corresponding pixel value of the target pixel in the three color channels of red, green and blue, and the RGB value of the peripheral associated pixel, calculated according to the contrast calculation formula the first difference between the RGB values of the eight pixels in the first peripheral layer and the target pixel, and the second difference between the RGB values of the sixteen pixels in the second peripheral layer and the target pixel, The first difference value and the second difference value are weighted and added to obtain the contrast of the target pixel point.
  • the contrast calculation formula and the above calculation method the contrast of each pixel in a certain superimposed image is obtained as the pixel contrast information of the superimposed image, and then the pixel contrast information of each superimposed image can be obtained sequentially according to the above method.
  • the formula for calculating the contrast ratio can be expressed by the following formula (3):
  • j 1 is the weighted value of the first difference
  • j 2 is the weighted value of the second difference
  • R u is the RGB value of the u-th pixel in the peripheral first layer
  • R v is the peripheral second layer.
  • R 0 is the RGB value of the target pixel.
  • the pixel contrast information of a superimposed image is obtained, and the contrast of each pixel is calculated according to the pixel contrast information.
  • Sort obtain multiple pixels in the sorting result that match the pixel dissolution ratio value and are ranked first as the dissolution pixels of the superimposed image, perform pixel dissolution on the customer image according to the dissolved pixels, and dissolve the non-dissolved pixels in the superimposed image. Pixels are deleted from the image to obtain the image contour information of the superimposed image.
  • the contour size information and the contour pixel information are extracted from the image contour information as feature information of each superimposed image.
  • the contour size information is the specific information of the contour size
  • the contour pixel information is the pixel value information of the image contour information.
  • the contour size information may include information such as contour length, contour width, and contour area
  • the contour pixel information may include information such as a chromaticity average value and a chromaticity variance value of the image contour information.
  • the chromaticity average value is an RGB average value obtained by averaging the RGB values of the pixels included in the image outline information
  • the chromaticity variance value is the RGB chromaticity of the pixels included in the image outline information.
  • the variance distribution value of is the variance distribution value of .
  • the data points are layered according to the characteristic information of each of the data points to obtain a plurality of data layers. According to the characteristic information of the data points, the data points can be layered to obtain a plurality of corresponding data layers, and each of the obtained data layers includes a plurality of data points.
  • step S130 includes sub-steps S131 and S132.
  • the types of multiple historical data corresponding to each data point can be determined, and the data points can be classified according to the type of each data point to obtain multiple data sets. For example, if the types of historical data include text and images, data points of the text type are classified into a data point set, and data points of the image type are classified into a data point set.
  • the data points contained in each data point set can be clustered according to the clustering rules and the characteristic information of the data points, then a sub-class obtained after clustering constitutes a data layer, and the clustering rule can be K-means clustering.
  • Class rules mean-shift clustering rules, Gaussian mixture model-based maximum expectation clustering rules or agglomerative hierarchical clustering rules, etc.
  • the data points contained in each data point set can be clustered to obtain a plurality of corresponding data layers.
  • the final conclusion can stand alone as a data layer.
  • S140 Generate a multi-layer related knowledge graph according to the connection relationship between the historical data and the feature information of the data points in each of the data layers.
  • a multi-layer relational knowledge graph is generated according to the connection relationship between the historical data and the feature information of the data points in each of the data layers.
  • the historical data information includes the connection relationship between the historical data, and a corresponding multi-layer relational knowledge graph can be generated according to the connection relationship of the historical data and the obtained multiple data layers.
  • step S140 includes sub-steps S141 , S142 and S143 .
  • a data point corresponds to a data node, and each data layer and the data nodes contained in the data layer are encoded to obtain data encoding information.
  • the data encoding information includes the data layer and the corresponding encoding value of the data node.
  • the coded values do not repeat each other, nor do the coded values of the data nodes.
  • the data nodes can be connected by an inverted index according to the connection relationship of the historical data. Specifically, the final conclusion is used as the starting point of the inverted index, and the multiple historical data contained in each event in the historical data information are concatenated with each other and each event points to A final conclusion is to perform inverted index joins on data nodes.
  • FIG. 2 is a schematic diagram of the effect of the information prediction method based on the multi-layer relational knowledge graph provided by the embodiment of the present application.
  • FIG. 2 is a partial diagram of the constructed multi-layer relational knowledge graph, and the data layer B1 and the data layer B2 are both basic layers.
  • the data layer L1.1 is the upper data layer of the data layer B1
  • A, B, C and D are the data nodes included in the data layer B1
  • the encoding values for encoding the data layer and data nodes in Figure 2 are only An example of an encoding method; the effect of performing an inverted index connection on data nodes is shown in Figure 2.
  • the newly added data feature information corresponding to the newly added data information is acquired according to the feature extraction model.
  • the user can input new data information, and then the prediction result corresponding to the new data information can be obtained based on the constructed multi-layer relational knowledge graph. Specifically, the user can input one piece of new data information or multiple pieces of new data information.
  • the inputted new data information may be all text information, or all image information, and may also include both text information and image information, and any piece of newly added data information may be image information or text information.
  • each new data information input by the user is text information in turn. If the new data information is text information, the new data can be extracted according to the text encoding dictionary and the feature extraction neural network in the feature extraction model. Perform feature extraction on the information to obtain the newly added data feature information corresponding to the newly added data information; if the newly added data information is image information, feature extraction can be performed on the newly added data information according to the image feature extraction rules in the feature extraction model , to obtain the newly added data feature information corresponding to the newly added data information, and the specific manner of acquiring the newly added data feature information is the same as the specific manner of acquiring the feature information of the data point, which will not be repeated here.
  • the multi-layer related knowledge graph includes a data node matching the information of the newly added data.
  • the data nodes whose feature information in the multi-layer related knowledge graph satisfies the judgment condition can be obtained as the data node matching the newly added data information.
  • the judgment condition is whether the feature information of the data node matches the A specific condition for judging whether the data feature information is matched is added, wherein the judging condition includes a matching degree calculation formula and a matching degree threshold.
  • step S160 includes sub-steps S161 and S162.
  • the multi-layer associated knowledge graph matches the type of each piece of new data feature information to match multiple data nodes. Specifically, if the new data information is text information, it is determined that The data node whose type is text is the data node that matches the newly added data information; if the newly added data information is image information, it is determined that the data node whose type is image in the knowledge graph is the data node that matches the newly added data information.
  • the matching degree calculation formula the matching degree between the newly added data feature information and the corresponding type of each data node feature information is calculated. Specifically, the matching degree calculation formula can be expressed by formula (4):
  • U i is the dimension value of the ith dimension in the feature information of the newly added data
  • V i is the dimension value of the ith dimension in the feature information corresponding to a certain data node of the corresponding type
  • n is the value of the ith dimension in the feature information of the new data.
  • the number of dimensions of the newly added data feature information is equal to the number of dimensions of the corresponding feature information of the corresponding type of data node, and the value range of the calculated matching degree is [0, 1].
  • the calculated matching degree it can be judged whether the number of data nodes whose matching degree with the newly added data information in the multi-layer relational knowledge graph is greater than the matching degree threshold is greater than zero. If the matching degree between the newly added data information is greater than the matching degree threshold If the number of data nodes is greater than zero, the obtained judgment result is that the multi-layer associated knowledge graph contains data nodes that match the newly added data information; otherwise, the obtained judgment result is that the multi-layer associated knowledge graph does not contain and newly added data nodes. The data information matches the data node.
  • the multi-layer relational knowledge graph includes data nodes that match the newly added data information, obtain the index connection information that matches the data nodes in the multi-layer relational knowledge graph as a corresponding prediction result .
  • the index connection information that matches the corresponding data nodes in the knowledge graph is obtained as the corresponding prediction result, and the prediction result can be used for the newly added data.
  • the development trend after the information is predicted, and the final conclusion pointed to by the index connection information in the prediction result is the prediction conclusion corresponding to the newly added data information.
  • the index connection information of the data node matching the newly added data information is obtained correspondingly as the prediction result;
  • the index connection information of the data node matching the data information is added, and the index connection information is deduplicated to obtain the final prediction result.
  • the new data information is text information
  • the new data feature information corresponding to the new data information is obtained, and it is judged that a multi-layer related knowledge graph is obtained.
  • the data node X in the data node X matches the newly added data feature information, the downstream of the data node X is connected to the data node Y and the data node Y points to the final conclusion of "cancer recurrence", and the index connection information of the data node X is obtained as the prediction result, then
  • the index connection information is the connection relationship between the data node X and the data node Y and the connection information of the final conclusion of "cancer recurrence" that the data node Y ultimately points to.
  • the prediction result is the prediction that the disease description information currently input by the user will follow the This possible connection path of "data node X->data node Y->cancer recurrence" in the knowledge graph is developed.
  • the new data information of the disease description information can be obtained according to the above method. Increase the data feature information and detect the new data feature information of the image, and judge that the data node P in the multi-layer relational knowledge graph matches the new data characteristic information of the disease description information, and the data node Q in the multi-layer relational knowledge graph matches.
  • the downstream of the data node P and the downstream of the data node Q are connected to the data node O, and the data node O points to the final conclusion of "cancer does not recur", then the data node P and the data node are obtained respectively.
  • the index connection information of the data node Q the index connection information of the two data nodes is deduplicated as a corresponding prediction result.
  • the multi-layer relational knowledge graph does not contain data nodes that match the newly added data information, then according to the characteristic information of the data nodes in each of the data layers, the corresponding data is extracted from each of the data layers.
  • the hierarchical feature information of The associated knowledge graph includes a data layer that matches the newly added data information, and a corresponding data node is generated according to the newly added data information and added to the data layer that matches the newly added data information; if the The multi-layer relational knowledge graph does not include a data layer matching the newly added data information, and corresponding independent data nodes are generated according to the newly added data information and added to the multi-layer relational knowledge graph.
  • the multiple independent data nodes can be aggregated to generate a new data layer to accommodate the corresponding multiple independent data nodes, and the newly generated data layer can be added to the multi-layer association in the knowledge graph.
  • the hierarchical feature information may be the average value corresponding to the feature information of the data nodes in the data layer. If the multi-layer relational knowledge graph contains a data layer that matches the newly added data information, the data corresponding to the newly added data information will be generated. The node is added to the data layer, and the newly generated data node is an independent data node in the data layer.
  • the independent data nodes are shown as data node K and data node H in FIG. 2 .
  • the data nodes matching the new data information are obtained in the above-mentioned manner, or the data nodes corresponding to the new data information are generated and added to the multi-layer related knowledge graph, and according to the new data information and The connection relationship between the newly-added data information entered last time, and the inverted index connection is performed between the newly generated data node and the last generated data node.
  • the method further includes: uploading the prediction result to the blockchain for storage.
  • the prediction result is uploaded to the blockchain for storage, and corresponding summary information is obtained based on the prediction result.
  • the summary information is obtained by hashing the prediction result, for example, by using the sha256 algorithm.
  • Uploading summary information to the blockchain ensures its security and fairness and transparency to users.
  • the user equipment can download the summary information from the blockchain in order to verify whether the prediction result has been tampered with.
  • the blockchain referred to in this example is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information to verify its Validity of information (anti-counterfeiting) and generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
  • the technical methods in this application can be applied to smart government affairs/smart city management/smart community/smart security/smart logistics/smart medical care/smart education/smart environmental protection/smart transportation and other application scenarios including intelligent prediction of information based on multi-layer associated knowledge graphs in order to promote the construction of smart cities.
  • the historical data information is aggregated to obtain multiple data points, the feature information corresponding to each data point is extracted, and the data points are stratified to obtain multiple data points.
  • a data layer is generated, according to the connection relationship between the historical data and the feature information of the data points in each data layer, the corresponding multi-layer relational knowledge graph is generated, and the newly-added characteristic data information corresponding to the newly-added data information is obtained.
  • the knowledge graph contains data nodes that match the newly added data information, and the index connection information corresponding to the data nodes is obtained as the prediction result.
  • the embodiment of the present application also provides an information prediction device 100 based on a multi-layer associated knowledge graph, the information prediction device based on a multi-layer associated knowledge graph can be configured in a user terminal or a management server, and the information based on the multi-layer associated knowledge graph
  • the prediction apparatus is configured to execute any of the foregoing embodiments of the information prediction method based on the multi-layer associated knowledge graph.
  • FIG. 9 is a schematic block diagram of an information prediction apparatus based on a multi-layer associated knowledge graph provided by an embodiment of the present application.
  • the information prediction apparatus 100 based on the multi-layer relational knowledge graph includes a historical data aggregation unit 110, a feature information extraction unit 120, a data layer acquisition unit 130, a knowledge graph generation unit 140, and a newly added data feature information acquisition unit 150 , a judging unit 160 and a prediction result obtaining unit 170 .
  • the historical data aggregation unit 110 is configured to aggregate the historical data included in the pre-stored historical data information according to a preset aggregation rule to obtain a plurality of corresponding data points.
  • the historical data aggregation unit 110 includes subunits: a historical data judging unit, for judging whether each of the historical data is text information; a keyword matching unit, for if the historical data is is text information, and obtains a plurality of keywords matching the historical data from the keyword set; the first aggregation unit is used for aggregating a plurality of the historical data containing the same keyword to obtain a corresponding data point; a similarity calculation unit, for calculating the similarity between the historical data according to the similarity calculation formula if the historical data is not text information; a second aggregation unit for A plurality of pieces of the historical data whose similarity is greater than the similarity threshold are aggregated to obtain a corresponding data point.
  • the feature information extraction unit 120 is configured to extract corresponding feature information from each of the data points according to a preset feature extraction model and historical data corresponding to each of the data points.
  • the feature information extraction unit 120 includes subunits: a data point judgment unit, used for judging whether the historical data corresponding to the data points are all text information; a coding information acquisition unit, used for The historical data corresponding to the data points are all text information, and the multiple keywords corresponding to each of the data points are converted according to the text encoding dictionary to obtain the encoding information of each of the data points; the feature information acquisition unit, for inputting the encoded information into the feature extraction neural network for calculation, to obtain feature information of each of the data points; an image feature extraction unit, for if the historical data corresponding to the data points are not all text information, The feature information corresponding to the data point is extracted from the historical data corresponding to the data point according to the image feature extraction rule.
  • the image feature extraction unit includes subunits: a superimposed image acquisition unit for superimposing historical data corresponding to each data point to obtain a superimposed image corresponding to each data point; pixel contrast information acquisition a unit, configured to calculate and obtain the pixel contrast information of each of the superimposed images according to the contrast calculation formula; an image contour information acquisition unit, used to calculate the pixel contrast information of each of the superimposed images and the dissolution ratio value for each of the superimposed images.
  • the image contour information is obtained by pixel dissolving the superimposed image; the contour feature information acquisition unit is configured to extract corresponding contour size information and contour pixel information from the image contour information of each of the superimposed images as the feature information.
  • the data layer obtaining unit 130 is configured to layer the data points according to the characteristic information of each data point to obtain multiple data layers.
  • the data layer acquiring unit 130 includes a sub-unit: a data point classification unit, configured to perform an analysis on each of the data points according to the type of the historical data and the historical data corresponding to each of the data points. A plurality of data point sets are obtained by classification; the data point clustering unit is used for clustering the data points contained in each of the data point sets according to the preset clustering rules and the characteristic information of the data points to obtain corresponding data points. Multiple data layers.
  • the knowledge graph generating unit 140 is configured to generate a multi-layer related knowledge graph according to the connection relationship between the historical data and the feature information of the data points in each of the data layers.
  • the knowledge graph generating unit 140 includes subunits: a data node generating unit, for generating a data node corresponding to each of the data points; The data layer and the data nodes included in each of the data layers are encoded to obtain data encoding information; an inverted index connection unit is used to invert the data nodes and the data layer according to the connection relationship of the historical data Index connection to obtain the corresponding multi-layer relational knowledge graph.
  • the newly added data feature information acquiring unit 150 is configured to acquire, according to the feature extraction model, newly added data feature information corresponding to the newly added data information if newly added data information input by the user is received.
  • the judging unit 160 is configured to judge whether the multi-layer related knowledge graph includes data nodes matching the newly added data information according to the preset judgment condition and the newly added data feature information.
  • the judging unit 160 includes a sub-unit: a matching degree calculation unit, configured to compare the difference between the characteristic information of the newly added data and the characteristic information of each of the data nodes according to the calculation formula of the matching degree. The matching degree is calculated; the judgment result acquisition unit is used to judge whether the number of data nodes whose matching degree is greater than the matching degree threshold is greater than zero, so as to obtain whether the multi-layer related knowledge graph contains information related to the newly added data. The judgment result of the matching data node.
  • the prediction result obtaining unit 170 is configured to obtain the index connection that matches the data node in the multi-layer associated knowledge graph if the multi-layer associated knowledge graph includes data nodes that match the newly added data information information as the corresponding prediction result.
  • the above-mentioned information prediction method based on a multi-layer relational knowledge graph is applied to the information prediction apparatus based on the multi-layer relational knowledge graph provided by the embodiment of the present application, and the historical data information is aggregated to obtain a plurality of data points, and the corresponding data points of each data point are extracted.
  • Feature information and stratify data points to obtain multiple data layers generate corresponding multi-layer related knowledge maps according to the connection relationship between historical data and the feature information of data points in each data layer, and obtain corresponding new data information
  • the newly added feature data information of if the multi-layer relational knowledge graph contains data nodes that match the newly added data information, the index connection information corresponding to the data nodes is obtained as the prediction result.
  • a multi-layer relational knowledge graph including multiple data layers and multiple data nodes is constructed based on historical data information, and prediction results corresponding to the newly added data information are obtained based on the multi-layer relational knowledge graph. Make trend forecasts.
  • the above-mentioned information prediction method based on a multi-layer relational knowledge graph can be implemented in the form of a computer program, and the computer program can be executed on a computer device as shown in FIG. 10 .
  • FIG. 10 is a schematic block diagram of a computer device provided by an embodiment of the present application.
  • the computer device may be a user terminal or a management server for executing an information prediction method based on a multi-layer associated knowledge graph to perform intelligent information prediction based on a multi-layer associated knowledge graph.
  • the computer device 500 includes a processor 502 , a memory and a network interface 505 connected through a system bus 501 , wherein the memory may include a storage medium 503 and an internal memory 504 .
  • the storage medium 503 can store an operating system 5031 and a computer program 5032 .
  • the processor 502 can execute the information prediction method based on the multi-layer relational knowledge graph, wherein the storage medium 503 can be a volatile storage medium or a non-volatile storage medium.
  • the processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500 .
  • the internal memory 504 provides an environment for running the computer program 5032 in the storage medium 503.
  • the processor 502 can execute the information prediction method based on the multi-layer relational knowledge graph.
  • the network interface 505 is used for network communication, such as providing transmission of data information.
  • the network interface 505 is used for network communication, such as providing transmission of data information.
  • FIG. 10 is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied.
  • the specific computer device 500 may include more or fewer components than shown, or combine certain components, or have a different arrangement of components.
  • the processor 502 is configured to run the computer program 5032 stored in the memory, so as to realize the corresponding functions in the above-mentioned information prediction method based on the multi-layer relational knowledge graph.
  • the embodiment of the computer device shown in FIG. 10 does not constitute a limitation on the specific structure of the computer device. Either some components are combined, or different component arrangements.
  • the computer device may only include a memory and a processor.
  • the structures and functions of the memory and the processor are the same as those of the embodiment shown in FIG. 10 , which will not be repeated here.
  • the processor 502 may be a central processing unit (Central Processing Unit, CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor can be a microprocessor or the processor can also be any conventional processor or the like.
  • a computer-readable storage medium may be a volatile or non-volatile computer-readable storage medium, and may also be a volatile computer-readable storage medium.
  • the computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the above-mentioned information prediction method based on a multi-layer relational knowledge graph is implemented.
  • the disclosed apparatus, apparatus and method may be implemented in other manners.
  • the apparatus embodiments described above are only illustrative.
  • the division of the units is only logical function division.
  • there may be other division methods, or units with the same function may be grouped into one Units, such as multiple units or components, may be combined or may be integrated into another system, or some features may be omitted, or not implemented.
  • the shown or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may also be electrical, mechanical or other forms of connection.
  • the units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solutions of the embodiments of the present application.
  • each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.
  • the above-mentioned integrated units may be implemented in the form of hardware, or may be implemented in the form of software functional units.
  • the integrated unit if implemented in the form of a software functional unit and sold or used as an independent product, may be stored in a computer-readable storage medium.
  • the read storage medium includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.
  • the aforementioned computer-readable storage medium includes: U disk, mobile hard disk, Read-Only Memory (ROM, Read-Only Memory), magnetic disk or optical disk and other media that can store program codes.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Library & Information Science (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种基于多层关联知识图谱的信息预测方法、装置、设备及介质，方法包括：对历史数据信息进行聚合得到多个数据点（S110），提取每一数据点对应的特征信息并对数据点进行分层得到多个数据层（S130），根据历史数据之间的连接关系及每一数据层中数据点的特征信息生成对应的多层关联知识图谱（S140），获取与新增数据信息对应的新增特征数据信息（S150），若多层关联知识图谱包含与新增数据信息相匹配的数据节点，获取数据节点对应的索引连接信息作为预测结果（S170）。该方法属于知识图谱技术且还涉及区块链技术，基于历史数据信息构建得到包含多个数据层及多个数据节点的多层关联知识图谱，并基于多层关联知识图谱获取与新增数据信息对应的预测结果，可准确地对信息进行趋势预测。

Description

基于多层关联知识图谱的信息预测方法、装置、设备及介质
本申请要求于2021年04月15日提交中国专利局、申请号为202110406359.9，发明名称为“基于多层关联知识图谱的信息预测方法、装置、设备及介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及知识图谱技术领域，属于智慧城市中基于多层关联知识图谱进行信息智能预测的应用场景，尤其涉及一种基于多层关联知识图谱的信息预测方法、装置及设备。
背景技术
随着信息技术的快速发展，基于信息之间的关联关系对海量信息进行处理的技术得到了越来越多的应用，可基于海量信息构建得到知识图谱，并基于知识图谱获取与相应信息相匹配的关联信息。然而发明人发现，现有的知识图谱仅能适用于对信息进行存储及关联匹配，而无法基于信息之间的联系对之后的趋势进行准确预测。因此，现有技术中的知识图谱存在无法对信息进行趋势预测的问题。
发明内容
本申请实施例提供了一种基于多层关联知识图谱的信息预测方法、装置、设备及介质，旨在解决现有技术中的知识图谱所存在的无法对信息进行趋势预测的问题。
第一方面，本申请实施例提供了一种基于多层关联知识图谱的信息预测方法，其中，所述方法包括：
根据预置的聚合规则对预存的历史数据信息中包含的历史数据进行聚合得到对应的多个数据点；
根据预置的特征提取模型及每一所述数据点对应的历史数据，从每一所述数据点中提取得到对应的特征信息；
根据每一所述数据点的特征信息对所述数据点进行分层得到多个数据层；
根据所述历史数据之间的连接关系及每一所述数据层中数据点的特征信息生成多层关联知识图谱；
若接收到用户输入的新增数据信息，根据所述特征提取模型获取与所述新增数据信息对应的新增数据特征信息；
根据预置的判断条件及所述新增数据特征信息对所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点进行判断；
若所述多层关联知识图谱中包含与所述新增数据信息相匹配的数据节点，获取所述多层关联知识图谱中与所述数据节点相匹配的索引连接信息作为对应的预测结果。
第二方面，本申请实施例提供了一种基于多层关联知识图谱的信息预测装置，其中，所述基于多层关联知识图谱的信息预测装置，包括：
历史数据聚合单元，用于根据预置的聚合规则对预存的历史数据信息中包含的历史数据进行聚合得到对应的多个数据点；
特征信息提取单元，用于根据预置的特征提取模型及每一所述数据点对应的历史数据，从每一所述数据点中提取得到对应的特征信息；
数据层获取单元，用于根据每一所述数据点的特征信息对所述数据点进行分层得到多个数据层；
知识图谱生成单元，用于根据所述历史数据之间的连接关系及每一所述数据层中数据点的特征信息生成多层关联知识图谱；
新增数据特征信息获取单元，用于若接收到用户输入的新增数据信息，根据所述特征提取模型获取与所述新增数据信息对应的新增数据特征信息；
判断单元，用于根据预置的判断条件及所述新增数据特征信息对所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点进行判断；
预测结果获取单元，用于若所述多层关联知识图谱中包含与所述新增数据信息相匹配的数据节点，获取所述多层关联知识图谱中与所述数据节点相匹配的索引连接信息作为对应的预测结果。
第三方面，本申请实施例又提供了一种计算机设备，其中，所述计算机设备包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序以实现上述第一方面所述的基于多层关联知识图谱的信息预测方法。
第四方面，本申请实施例还提供了一种计算机可读存储介质，其中，所述计算机可读存储介质存储有计算机程序，当所述计算机程序被处理器执行时实现上述第一方面所述的基于多层关联知识图谱的信息预测方法。
本申请实施例提供了一种基于多层关联知识图谱的信息预测方法、装置、设备及介质。对历史数据信息进行聚合得到多个数据点，提取每一数据点对应的特征信息并对数据点进行分层得到多个数据层，根据历史数据之间的连接关系及每一数据层中数据点的特征信息生成对应的多层关联知识图谱，获取与新增数据信息对应的新增特征数据信息，若多层关联知识图谱包含与新增数据信息相匹配的数据节点，获取数据节点对应的索引连接信息作为预测结果。通过上述方法，基于历史数据信息构建得到包含多个数据层及多个数据节点的多层关联知识图谱，并基于多层关联知识图谱获取与新增数据信息对应的预测结果，可准确地对信息进行趋势预测。
附图说明
为了更清楚地说明本申请实施例技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本申请实施例提供的基于多层关联知识图谱的信息预测方法的流程示意图；
图2为本申请实施例提供的基于多层关联知识图谱的信息预测方法的效果示意图；
图3为本申请实施例提供的基于多层关联知识图谱的信息预测方法的子流程示意图；
图4为本申请实施例提供的基于多层关联知识图谱的信息预测方法的另一子流程示意图；
图5为本申请实施例提供的基于多层关联知识图谱的信息预测方法的另一子流程示意图；
图6为本申请实施例提供的基于多层关联知识图谱的信息预测方法的另一子流程示意图；
图7为本申请实施例提供的基于多层关联知识图谱的信息预测方法的另一子流程示意图；
图8为本申请实施例提供的基于多层关联知识图谱的信息预测方法的另一子流程示意图；
图9为本申请实施例提供的基于多层关联知识图谱的信息预测装置的示意性框图；
图10为本申请实施例提供的计算机设备的示意性框图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
应当理解，当在本说明书和所附权利要求书中使用时，术语“包括”和“包含”指示所描述特征、整体、步骤、操作、元素和/或组件的存在，但并不排除一个或多个其它特征、整体、步骤、操作、元素、组件和/或其集合的存在或添加。
还应当理解，在此本申请说明书中所使用的术语仅仅是出于描述特定实施例的目的而并不意在限制本申请。如在本申请说明书和所附权利要求书中所使用的那样，除非上下文清楚地指明其它情况，否则单数形式的“一”、“一个”及“该”意在包括复数形式。
还应当进一步理解，在本申请说明书和所附权利要求书中使用的术语“和/或”是指相关联列出的项中的一个或多个的任何组合以及所有可能组合，并且包括这些组合。
请参阅图1，图1是本申请实施例提供的基于多层关联知识图谱的信息预测方法的流程示意图；该基于多层关联知识图谱的信息预测方法应用于用户终端或管理服务器中，该基于多层关联知识图谱的信息预测方法通过安装于用户终端或管理服务器中的应用软件进行执行，用户终端即是对用户输入的新增数据信息进行智能预测的终端设备，例如台式电脑、笔记本电脑、平板电脑、手机等，管理服务器即是接收用户所输入的新增数据信息进行智能预测的服务器端，如企业或政府部门内构建的服务器。如图1所示，该方法包括步骤S110～S170。
S110、根据预置的聚合规则对预存的历史数据信息中包含的历史数据进行聚合得到对应的多个数据点。
根据预置的聚合规则对预存的历史数据信息中包含的历史数据进行聚合得到对应的多个数据点。历史数据信息可由多条历史数据组成，每一条历史数据可以是文本信息或图像信息，历史数据信息中包含多个同类型且相互独立的事件，每一事件可对应包含多条历史数据，同一事件包含的多条历史数据相互串联并指向一个最终结论，这也即是历史数据之间的连接关系，历史数据信息中的一个事件可仅包含文本信息，也可以仅包含图像信息，还可以同时包含文本信息及图像信息。
历史数据可以是气象资料的历史数据或癌症复发检测的历史数据等。例如，事件可以是下雨事件，则最终结论即为下雨或不下雨，在一个下雨事件的最终结论出现之前不同时间点所采集得到的信息即构成该事件的历史数据，如对下雨事件之前每隔一段时间记录得到的空气温度、湿度等信息即构成该事件的文字信息，对下雨事件之前每隔一段时间获取到的卫星云图即构成该事件的图像信息，则可根据时间顺序对采集得到的文字信息及图像信息进行串联连接，并指向下雨或不下雨的最终结论。
在构建知识图谱之前，需要对历史数据信息中包含的历史数据进行聚合整理，具体的，可通过聚合规则对历史数据进行聚合处理。其中，所述聚合规则包括关键词集合、相似度计算公式及相似度阈值。
在一实施例中，如图3所示，步骤S110包括子步骤S111、S112、S113、S114和S115。
S111、对每一所述历史数据是否为文本信息进行判断。
可首先判断历史数据是否为文本信息，若为文本信息即对均为文本信息的历史数据进行聚合处理，若不为文本信息，即可对均为图像信息的历史数据进行聚合处理。
S112、若所述历史数据为文本信息，从所述关键词集合中获取与历史数据相匹配的多个关键词。
关键词集合即包含与历史数据信息类型相同的多个关键词，可根据关联词集合获取与文本信息的历史数据相匹配的多个关键词。
例如，若历史数据信息类型为癌症复发检测，则对应的历史数据为一段病情描述信息，与该类型对应的关键词集合可包含关键词：积液、病灶、病变、水肿、白细胞偏高等。
可根据历史数据包含的文本内容，从关键词集合中获取与该文本内容相匹配的多个关键词，可从关键词集合中获取与每一历史数据相匹配的三个关键词，若与历史数据相匹配的关键词数量大于三个，则根据每一关键词在该历史数据中的出现频次，获取出现频次最高的三个关键词作为与该历史数据相匹配的三个关键词。
S113、对包含相同关键词的多条所述历史数据进行聚合，得到对应的一个数据点。
若两条历史数据或多条历史数据均包含相同关键词，则表明两条历史数据或多条历史数据均表达相同内容，可对两条或多条历史数据进行聚合得到一个数据点，则所得到的每一个数据点至少对应一条历史数据。
S114、若所述历史数据不为文本信息，根据所述相似度计算公式对所述历史数据之间的 相似度进行计算。
若历史数据不为文本信息，则历史数据为历史图像，历史图像的图像尺寸均相等，历史图像可以是彩色图像。例如，历史数据信息类型为癌症复发检测，则对应的历史数据可以是一张疾病部位的检测图像。
可通过相似度计算公式对两张历史图像之间的相似度进行计算，具体的，首先分别对两张历史图像进行灰度处理得到两张灰度图像，灰度值采用非负整数进行表示，像素对应灰度值的取值范围为[0，255]，灰度值为0则表示该像素点为黑色，灰度值为255则表示该像素点为白色，灰度值为其他数值则表明该像素点为介于白色与黑色之间的一个具体灰度。可以采用公式(1)计算得到每一灰度图像的图像方差值：
Figure PCTCN2021097105-appb-000001
任意一张灰度图像的图像方差值可表示为
Figure PCTCN2021097105-appb-000002
其中，
Figure PCTCN2021097105-appb-000003
为灰度图像第m行的方差值，Am为灰度图像第m行所有像素点的灰度平均值，B为灰度图像所有像素点的灰度平均值，N为所述灰度图像所包含的总行数；例如，
Figure PCTCN2021097105-appb-000004
为灰度图像F第m行的方差值，则图像方差值中包含灰度图像F逐行的方差值，Am为灰度图像F第m行所有像素点的灰度平均值，B为灰度图像F所有像素点的灰度平均值，N为灰度图像F所包含的总行数。其中，N也可以为灰度图像F所包含的总列数。
相似度计算公式可采用公式(2)进行表示：
Figure PCTCN2021097105-appb-000005
根据两张灰度图像的图像方差值，计算得到两张灰度图像之间的相似度Wσ，其中，
Figure PCTCN2021097105-appb-000006
为第一张灰度图像第m行的方差值，
Figure PCTCN2021097105-appb-000007
为第二张灰度图像第m行的方差值。
S115、对互相之间相似度大于所述相似度阈值的多条所述历史数据进行聚合，得到对应的一个数据点。
对两张灰度图像之间的相似度是否大于相似度阈值进行判断，若大于，也即表明与两张灰度图像对应的两条历史数据之间的相似度大于相似度阈值；若不大于，则表明对应的两条历史数据之间相似度不大于相似度阈值。可根据计算得到的相似度，对互相之间相似度大于相似度阈值的多条历史数据进行聚合，得到对应的一个数据点，则此时每一个数据点至少对应一条历史数据。
例如，若得到历史数据F A与历史数据F B之间相似度大于相似度阈值，则对两条历史数据进行聚合处理；若得到历史数据F A与历史数据F B之间相似度大于相似度阈值，且历史数据F B与历史数据F C之间相似度也大于相似度阈值，则对三条历史数据F A、F B及F C进行聚合处理。
S120、根据预置的特征提取模型及每一所述数据点对应的历史数据，从每一所述数据点中提取得到对应的特征信息。
根据预置的特征提取模型及每一所述数据点对应的历史数据，从每一所述数据点中提取得到对应的特征信息。特征提取模型即为对数据点的特征信息进行提取的具体模型，其中，所述特征提取模型包括文本编码词典、特征提取神经网络及图像特征提取规则，数据点的特征信息可采用特征向量的形式进行表示。
在一实施例中，如图4所示，步骤S120包括子步骤S121、S122、S123和S124。
S121、对所述数据点对应的历史数据是否均为文本信息进行判断；
任意一个数据点所对应的历史数据要么均为文本信息，要么均为图像信息，则可相对数据点对应的历史数据的类型进行判断，并根据判断结果对每一数据点的历史数据进行相应处理。
S122、若所述数据点对应的历史数据均为文本信息，根据所述文本编码词典对每一所述数据点对应的多个关键词进行转换，得到每一所述数据点的编码信息。
若某一数据点对应的历史数据均为文本信息，则该数据点对应的历史数据中均包含相同的多个关键词，则根据文本编码词典对该数据点对应的多个关键词进行转换，得到该数据点的编码信息。具体的，每一字符均可在文本编码词典中匹配到对应的一个特征编码，则可根据文本编码词典对每一数据点对应的多个关键词中的字符进行转换，将转换得到的多个特征编码进行顺序组合得到对应数据点的编码信息，转换得到的编码信息可采用大小为(1，R)的特征向量进行表示，则编码信息为一个1行R列的特征向量，R为预先设定的一个长度值，例如，可将R设定为12，则获取每一数据点对应的多个特征编码组合得到编码信息，若特征编码数据不足12，则最后几位数值采用“0”进行补齐。
S123、将所述编码信息输入所述特征提取神经网络进行计算，得到每一所述数据点的特征信息。
可将所得到的每一数据点的编码信息输入特征提取神经网络进行计算，得到每一数据点的特征信息。具体的，特征提取神经网络即为基于人工智能构建得到的神经网络，特征提取神经网络由一个输入层、多个中间层及一个输出层组成，输入层与中间层之间、中间层与其他中间层之间、中间层与输出层之间均通过关联公式进行关联，例如某一关联公式可表示为y＝c 1×x+c 2，c 1和c 2即为该关联公式中的参数值。输入层中包含的输入节点的数量与编码信息中向量维度数量相等，则编码信息中每一维度的向量值均与一个输入节点相对应。将一个数据点的编码信息输入特征提取神经网络进行计算，即可从特征提取神经网络的输出层获取相应的特征信息，特征信息可采用一个特征数组(R，T)进行表示，特征数组的大小为R行T列，且特征数组中每一数值均属于[0，1]这一取值范围。
S124、若所述数据点对应的历史数据不均为文本信息，根据所述图像特征提取规则从所述数据点对应的历史数据中提取得到与所述数据点对应的特征信息。
若某一数据点对应的历史数据均为图像信息，则可从该数据点对应的多张图像信息中提取得到对应的特征信息，其中，图像特征提取规则即为从多张图像信息中提取对应特征信息的具体规则，所述图像特征提取规则包括对比度计算公式及溶解比例值。
在一实施例中，如图5所示，步骤S124包括子步骤S1241、S1242、S1243和S1244。
S1241、对每一所述数据点对应的历史数据进行叠加得到每一数据点对应的叠加图像。
数据点对应的每一历史数据均为图像信息，且图像信息的尺寸均相同，则可对同一数据点对应的多张图像信息进行叠加，具体的，对多张图像信息每一像素点位置对应的像素值进行平均计算，也即是计算每一像素点的RGB值的平均值，RGB值中包含红、绿、蓝三个颜色通道分别对应的色度值，则需要分别计算红、绿、蓝三个通道分别对应的像素值的平均值，得到每一像素点的像素平均值，同一数据点包含的图像信息中每一像素点的像素平均值即组合成为一张叠加图像。
S1242、根据所述对比度计算公式计算得到每一所述叠加图像的像素对比度信息。
具体的，获取一张叠加图像并确定其中的一个像素点为目标像素点，可获取目标像素点外围第一层的八个像素点及外围第二层的十六个像素点作为该目标像素点的关联像素点，获取目标像素点的RGB值，GRB值包括目标像素点在红、绿、蓝三个颜色通道分别对应的像素值，以及外围的关联像素点的RGB值，根据对比度计算公式计算外围第一层的八个像素点的RGB值与目标像素点之间的第一差值，以及外围第二层的十六个像素点的RGB值与目标像素点之间的第二差值，并对第一差值及第二差值进行加权相加得到该目标像素点的对比度。根据对比度计算公式及上述计算方法获取某一张叠加图像中每一像素点的对比度作为该叠加图像的像素对比度信息，则可根据上述方法依次获取得到每一张叠加图像的像素对比度信息。
例如，对比度计算公式可采用以下公式(3)进行表示：
Figure PCTCN2021097105-appb-000008
其中，j 1为第一差值的加权值，j 2为第二差值的加权值，R u为外围第一层中的第u个像素 点的RGB值，R v为外围第二层中的第v个像素点的RGB值，R 0为目标像素点的RGB值。
S1243、根据每一所述叠加图像的像素对比度信息及所述溶解比例值对每一所述叠加图像进行像素溶解得到图像轮廓信息。
叠加图像中像素点的对比度越大，也即表明该像素点与周边像素点之间的差异度越大，获取某一叠加图像的像素对比度信息，根据像素对比度信息对每一像素点的对比度进行排序，获取排序结果中与像素溶解比例值相匹配且排序靠前的多个像素点作为该叠加图像的溶解像素点，根据溶解像素点对该客户图像进行像素溶解，将该叠加图像中非溶解像素点从图像中删除，得到该叠加图像的图像轮廓信息。
S1244、从每一所述叠加图像的图像轮廓信息中提取得到对应的轮廓尺寸信息及轮廓像素信息作为所述特征信息。
从图像轮廓信息中提取得到轮廓尺寸信息及轮廓像素信息，作为每一叠加图像的特征信息，轮廓尺寸信息即为轮廓外形尺寸的具体信息，轮廓像素信息即为图像轮廓信息的像素值信息。具体的，轮廓尺寸信息可以包括轮廓长度、轮廓宽度及轮廓面积等信息，轮廓像素信息可以包括图像轮廓信息的色度平均值、色度方差值等信息。具体的，色度平均值即为对图像轮廓信息包含的像素点的RGB值进行平均计算所得到的一个RGB平均值，色度方差值为图像轮廓信息中包含的像素点在RGB色度上的方差分布值。
S130、根据每一所述数据点的特征信息对所述数据点进行分层得到多个数据层。
根据每一所述数据点的特征信息对所述数据点进行分层得到多个数据层。根据数据点的特征信息可将数据点进行分层得到对应的多个数据层，则得到的每一数据层中均包含多个数据点。
在一实施例中，如图6所示，步骤S130包括子步骤S131和S132。
S131、根据所述历史数据的类型及每一所述数据点对应的历史数据对每一所述数据点进行分类得到多个数据点集。
首先可根据历史数据的类型，确定每一数据点对应的多个历史数据的类型，并根据每一数据点的类型对数据点进行分类得到多个数据集集。例如，历史数据的类型包括文本及图像，则文本类型的数据点分类至一个数据点集，图像类型的数据点分类至一个数据点集。
S132、根据预置的聚类规则及所述数据点的特征信息，对每一所述数据点集包含的数据点进行聚类得到对应的多个数据层。
可根据聚类规则及数据点的特征信息对每一数据点集包含的数据点进行聚类，则聚类后所得到的一个小类即构成一个数据层，聚类规则可以是K-means聚类规则、均值漂移聚类规则、基于高斯混合模型的最大期望聚类规则或凝聚层次聚类规则等。则每一数据点集包含的数据点均可聚类得到对应的多个数据层。此外，最终结论可单独作为一个数据层。
S140、根据所述历史数据之间的连接关系及每一所述数据层中数据点的特征信息生成多层关联知识图谱。
根据所述历史数据之间的连接关系及每一所述数据层中数据点的特征信息生成多层关联知识图谱。历史数据信息中包含历史数据之间的连接关系，可根据历史数据的连接关系及所得得到的多个数据层生成对应的多层关联知识图谱。
在一实施例中，如图7所示，步骤S140包括子步骤S141、S142和S143。
S141、生成与每一所述数据点对应的数据节点；S142、分别对每一所述数据层及每一所述数据层包含的数据节点进行编码得到数据编码信息。
一个数据点对应生成一个数据节点，并对每一数据层及数据层内包含的数据节点进行编码，得到数据编码信息，则数据编码信息中包括数据层及数据节点对应的编码值，数据层的编码值互不重复，数据节点的编码值也互不重复。
S143、根据所述历史数据的连接关系对所述数据节点及所述数据层进行倒排索引连接， 得到对应的多层关联知识图谱。
可根据历史数据的连接关系对数据节点进行倒排索引连接，具体的，以最终结论作为倒排索引的起点，根据历史数据信息中每一事件包含的多条历史数据相互串联以及每一事件指向一个最终结论，对数据节点进行倒排索引连接。
图2为本申请实施例提供的基于多层关联知识图谱的信息预测方法的效果示意图，图2即为所构建的多层关联知识图谱的局部图，数据层B1及数据层B2均为基础层，数据层L1.1为数据层B1的上一层数据层，A、B、C及D均为数据层B1包含的数据节点，图2中对数据层及数据节点进行编码的编码值仅为一种编码方式的示例；对数据节点进行倒排索引连接的效果如图2所示。
S150、若接收到用户输入的新增数据信息，根据所述特征提取模型获取与所述新增数据信息对应的新增数据特征信息。
若接收到用户输入的新增数据信息，根据所述特征提取模型获取与所述新增数据信息对应的新增数据特征信息。用户可输入新增数据信息，则可基于构建的多层关联知识图谱获取与新增数据信息对应的预测结果，具体的，用户可输入一条新增数据信息或输入多条新增数据信息，用户输入的新增数据信息可以均为文本信息，或者均为图像信息，还可以即包含文本信息也包含图像信息，任意一条新增数据信息可以是图像信息或文本信息。
可先对用户输入的每一条新增数据信息是否为文本信息依次进行判断，若新增数据信息为文本信息，则可根据特征提取模型中的文本编码词典及特征提取神经网络对该新增数据信息进行特征提取，得到与该新增数据信息对应的新增数据特征信息；若新增数据信息为图像信息，则可根据特征提取模型中的图像特征提取规则对该新增数据信息进行特征提取，得到与该新增数据信息对应的新增数据特征信息，获取新增数据特征信息的具体方式与获取数据点的特征信息的具体方式相同，在此不作赘述。
S160、根据预置的判断条件及所述新增数据特征信息对所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点进行判断。
根据预置的判断条件及所述新增数据特征信息对所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点进行判断。可基于所得到的新增数据特征信息，获取多层关联知识图谱中特征信息满足判断条件的数据节点作为与新增数据信息相匹配的数据节点，判断条件即为对数据节点的特征信息是否与新增数据特征信息之间是否相匹配进行判断的具体条件，其中，所述判断条件包括匹配度计算公式及匹配度阈值。
在一实施例中，如图8所示，步骤S160包括子步骤S161和S162。
S161、根据所述匹配度计算公式对所述新增数据特征信息与每一所述数据节点的特征信息之间的匹配度进行计算。
根据每一条新增数据信息的类型，确定多层关联知识图谱与每一条新增数据特征信息的类型相匹配多个数据节点，具体的，若新增数据信息为文本信息，则确定知识图谱中类型为文本的数据节点为与新增数据信息相匹配的数据节点；若新增数据信息为图像信息，则确定知识图谱中类型为图像的数据节点为与新增数据信息相匹配的数据节点。根据匹配度计算公式对新增数据特征信息与相应类型的每一数据节点特征信息之间的匹配度进行计算，具体的，匹配度计算公式可采用公式(4)进行表示：
Figure PCTCN2021097105-appb-000009
其中，U i为新增数据特征信息中第i个维度的维度值，V i为相应类型的某一数据节点对应特征信息中第i个维度的维度值，n为新增数据特征信息所包含的维度总数，新增数据特征信息的维度数与相应类型的数据节点对应特征信息的维度数相等，计算得到的匹配度的取值范围为[0，1]。
S162、判断匹配度大于所述匹配度阈值的数据节点的数量是否大于零，以得到所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点的判断结果。
可根据计算得到的匹配度，判断多层关联知识图谱中与新增数据信息之间匹配度大于匹配度阈值的数据节点的数量是否大于零，若新增数据信息之间匹配度大于匹配度阈值的数据节点的数量大于零，则得到的判断结果为多层关联知识图谱中包含与新增数据信息相匹配的数据节点；否则，得到的判断结果为多层关联知识图谱中不包含与新增数据信息相匹配的数据节点。
S170、若所述多层关联知识图谱中包含与所述新增数据信息相匹配的数据节点，获取所述多层关联知识图谱中与所述数据节点相匹配的索引连接信息作为对应的预测结果。
若判断结果为多层关联知识图谱中包含与新增数据信息相匹配的数据节点，则获取知识图谱中与相应数据节点相匹配的索引连接信息作为对应预测结果，预测结果即可对新增数据信息之后的发展趋势进行预测，预测结果中索引连接信息指向的最终结论即为新增数据信息对应的预测结论。具体的，若用户仅输入一条新增数据信息，则对应获取与该新增数据信息相匹配的数据节点的索引连接信息作为预测结果；若用户输入多条新增数据信息，可获取与每一新增数据信息相匹配的数据节点的索引连接信息，并对索引连接信息进行去重处理得到最终预测结果。
例如，用户输入的某一条新增数据信息为一段病情描述信息，则该新增数据信息为文本信息，获取与该新增数据信息对应的新增数据特征信息，并判断得到多层关联知识图谱中的数据节点X与新增数据特征信息相匹配，数据节点X的下游连接数据节点Y且数据节点Y指向“癌症复发”这一最终结论，获取数据节点X的索引连接信息作为预测结果，则索引连接信息即为数据节点X与数据节点Y的连接关系以及数据节点Y最终指向的“癌症复发”这一最终结论的连接信息，预测结果也即是预测用户当前输入的病情描述信息会沿着知识图谱中“数据节点X->数据节点Y->癌症复发”的这一可能的连接路径进行发展。若用户同时输入两条新增数据信息，其中一条新增数据信息为一段病情描述信息，另一条新增数据信息为一张疾病部位的检测图像，则可根据上述方法对应获取病情描述信息的新增数据特征信息及检测图像的新增数据特征信息，并判断得到多层关联知识图谱中的数据节点P与病情描述信息的新增数据特征信息相匹配、多层关联知识图谱中的数据节点Q与检测图像的新增数据特征信息相匹，数据节点P的下游及数据节点Q的下游均连接数据节点O，数据节点O指向“癌症不复发”这一最终结论，则分别获取数据节点P及数据节点Q的索引连接信息，对两个数据节点的索引连接信息进行去重后作为对应的预测结果。
若所述多层关联知识图谱中不包含与所述新增数据信息相匹配的数据节点，则根据每一所述数据层中数据节点的特征信息，从每一所述数据层中提取得到对应的分层特征信息；根据所述判断条件及所述新增数据特征信息对所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据层进行判断；若所述多层关联知识图谱中包含与所述新增数据信息相匹配的数据层，根据所述新增数据信息生成对应的数据节点并添加至与所述新增数据信息相匹配的数据层中；若所述多层关联知识图谱中不包含与所述新增数据信息相匹配的数据层，根据所述新增数据信息生成对应的独立数据节点并添加至所述多层关联知识图谱中。若多个独立数据节点之间满足相应聚合条件，则可对多个独立数据节点进行聚合，以生成新的数据层容纳对应的多个独立数据节点，并将新生成的数据层添加至多层关联知识图谱中。
其中，分层特征信息可以是数据层中数据节点的特征信息对应平均值，若多层关联知识图谱中包含与新增数据信息相匹配的数据层，则将生成与新增数据信息对应的数据节点并添加至该数据层中，新生成的数据节点即为数据层中独立的数据节点，独立的数据节点如图2中数据节点K及数据节点H所示。当再次输入相应新增数据信息，则按上述方式获取与新增数据信息相匹配的数据节点或生成与新增数据信息对应的数据节点添加至多层关联知识图谱 中，并根据新增数据信息与上次输入的新增数据信息之间的连接关系，将新生成的数据节点与上次生成的数据节点进行倒排索引连接。
在一实施例中，步骤S170之后还包括：将所述预测结果上传至区块链进行存储。
将所述预测结果上传至区块链进行存储，基于预测结果得到对应的摘要信息，具体来说，摘要信息由预测结果进行散列处理得到，比如利用sha256算法处理得到。将摘要信息上传至区块链可保证其安全性和对用户的公正透明性。用户设备可以从区块链中下载得该摘要信息，以便查证预测结果是否被篡改。本示例所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain)，本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
本申请中的技术方法可应用于智慧政务/智慧城管/智慧社区/智慧安防/智慧物流/智慧医疗/智慧教育/智慧环保/智慧交通等包含基于多层关联知识图谱进行信息智能预测的应用场景中，从而推动智慧城市的建设。
在本申请实施例所提供的基于多层关联知识图谱的信息预测方法中，对历史数据信息进行聚合得到多个数据点，提取每一数据点对应的特征信息并对数据点进行分层得到多个数据层，根据历史数据之间的连接关系及每一数据层中数据点的特征信息生成对应的多层关联知识图谱，获取与新增数据信息对应的新增特征数据信息，若多层关联知识图谱包含与新增数据信息相匹配的数据节点，获取数据节点对应的索引连接信息作为预测结果。通过上述方法，基于历史数据信息构建得到包含多个数据层及多个数据节点的多层关联知识图谱，并基于多层关联知识图谱获取与新增数据信息对应的预测结果，可准确地对信息进行趋势预测。
本申请实施例还提供一种基于多层关联知识图谱的信息预测装置100，该基于多层关联知识图谱的信息预测装置可配置于用户终端或管理服务器中，该基于多层关联知识图谱的信息预测装置用于执行前述的基于多层关联知识图谱的信息预测方法的任一实施例。具体地，请参阅图9，图9为本申请实施例提供的基于多层关联知识图谱的信息预测装置的示意性框图。
如图9所示，该基于多层关联知识图谱的信息预测装置100包括历史数据聚合单元110、特征信息提取单元120、数据层获取单元130、知识图谱生成单元140、新增数据特征信息获取单元150、判断单元160和预测结果获取单元170。
历史数据聚合单元110，用于根据预置的聚合规则对预存的历史数据信息中包含的历史数据进行聚合得到对应的多个数据点。
在一实施例中，所述历史数据聚合单元110包括子单元：历史数据判断单元，用于对每一所述历史数据是否为文本信息进行判断；关键词匹配单元，用于若所述历史数据为文本信息，从所述关键词集合中获取与历史数据相匹配的多个关键词；第一聚合单元，用于对包含相同关键词的多条所述历史数据进行聚合，得到对应的一个数据点；相似度计算单元，用于若所述历史数据不为文本信息，根据所述相似度计算公式对所述历史数据之间的相似度进行计算；第二聚合单元，用于对互相之间相似度大于所述相似度阈值的多条所述历史数据进行聚合，得到对应的一个数据点。
特征信息提取单元120，用于根据预置的特征提取模型及每一所述数据点对应的历史数据，从每一所述数据点中提取得到对应的特征信息。
在一实施例中，所述特征信息提取单元120包括子单元：数据点判断单元，用于对所述数据点对应的历史数据是否均为文本信息进行判断；编码信息获取单元，用于若所述数据点对应的历史数据均为文本信息，根据所述文本编码词典对每一所述数据点对应的多个关键词进行转换，得到每一所述数据点的编码信息；特征信息获取单元，用于将所述编码信息输入 所述特征提取神经网络进行计算，得到每一所述数据点的特征信息；图像特征提取单元，用于若所述数据点对应的历史数据不均为文本信息，根据所述图像特征提取规则从所述数据点对应的历史数据中提取得到与所述数据点对应的特征信息。
在一实施例中，所述图像特征提取单元包括子单元：叠加图像获取单元，用于对每一所述数据点对应的历史数据进行叠加得到每一数据点对应的叠加图像；像素对比度信息获取单元，用于根据所述对比度计算公式计算得到每一所述叠加图像的像素对比度信息；图像轮廓信息获取单元，用于根据每一所述叠加图像的像素对比度信息及所述溶解比例值对每一所述叠加图像进行像素溶解得到图像轮廓信息；轮廓特征信息获取单元，用于从每一所述叠加图像的图像轮廓信息中提取得到对应的轮廓尺寸信息及轮廓像素信息作为所述特征信息。
数据层获取单元130，用于根据每一所述数据点的特征信息对所述数据点进行分层得到多个数据层。
在一实施例中，所述数据层获取单元130包括子单元：数据点分类单元，用于根据所述历史数据的类型及每一所述数据点对应的历史数据对每一所述数据点进行分类得到多个数据点集；数据点聚类单元，用于根据预置的聚类规则及所述数据点的特征信息，对每一所述数据点集包含的数据点进行聚类得到对应的多个数据层。
知识图谱生成单元140，用于根据所述历史数据之间的连接关系及每一所述数据层中数据点的特征信息生成多层关联知识图谱。
在一实施例中，所述知识图谱生成单元140包括子单元：数据节点生成单元，用于生成与每一所述数据点对应的数据节点；数据编码信息获取单元，用于分别对每一所述数据层及每一所述数据层包含的数据节点进行编码得到数据编码信息；倒排索引连接单元，用于根据所述历史数据的连接关系对所述数据节点及所述数据层进行倒排索引连接，得到对应的多层关联知识图谱。
新增数据特征信息获取单元150，用于若接收到用户输入的新增数据信息，根据所述特征提取模型获取与所述新增数据信息对应的新增数据特征信息。
判断单元160，用于根据预置的判断条件及所述新增数据特征信息对所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点进行判断。
在一实施例中，所述判断单元160包括子单元：匹配度计算单元，用于根据所述匹配度计算公式对所述新增数据特征信息与每一所述数据节点的特征信息之间的匹配度进行计算；判断结果获取单元，用于判断匹配度大于所述匹配度阈值的数据节点的数量是否大于零，以得到所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点的判断结果。
预测结果获取单元170，用于若所述多层关联知识图谱中包含与所述新增数据信息相匹配的数据节点，获取所述多层关联知识图谱中与所述数据节点相匹配的索引连接信息作为对应的预测结果。
在本申请实施例所提供的基于多层关联知识图谱的信息预测装置应用上述基于多层关联知识图谱的信息预测方法，对历史数据信息进行聚合得到多个数据点，提取每一数据点对应的特征信息并对数据点进行分层得到多个数据层，根据历史数据之间的连接关系及每一数据层中数据点的特征信息生成对应的多层关联知识图谱，获取与新增数据信息对应的新增特征数据信息，若多层关联知识图谱包含与新增数据信息相匹配的数据节点，获取数据节点对应的索引连接信息作为预测结果。通过上述方法，基于历史数据信息构建得到包含多个数据层及多个数据节点的多层关联知识图谱，并基于多层关联知识图谱获取与新增数据信息对应的预测结果，可准确地对信息进行趋势预测。
上述基于多层关联知识图谱的信息预测方法可以实现为计算机程序的形式，该计算机程序可以在如图10所示的计算机设备上运行。
请参阅图10，图10是本申请实施例提供的计算机设备的示意性框图。该计算机设备可 以是用于执行基于多层关联知识图谱的信息预测方法以基于多层关联知识图谱进行信息智能预测的用户终端或管理服务器。
参阅图10，该计算机设备500包括通过系统总线501连接的处理器502、存储器和网络接口505，其中，存储器可以包括存储介质503和内存储器504。
该存储介质503可存储操作系统5031和计算机程序5032。该计算机程序5032被执行时，可使得处理器502执行基于多层关联知识图谱的信息预测方法，其中，存储介质503可以为易失性的存储介质或非易失性的存储介质。
该处理器502用于提供计算和控制能力，支撑整个计算机设备500的运行。
该内存储器504为存储介质503中的计算机程序5032的运行提供环境，该计算机程序5032被处理器502执行时，可使得处理器502执行基于多层关联知识图谱的信息预测方法。
该网络接口505用于进行网络通信，如提供数据信息的传输等。本领域技术人员可以理解，图10中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备500的限定，具体的计算机设备500可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
其中，所述处理器502用于运行存储在存储器中的计算机程序5032，以实现上述的基于多层关联知识图谱的信息预测方法中对应的功能。
本领域技术人员可以理解，图10中示出的计算机设备的实施例并不构成对计算机设备具体构成的限定，在其他实施例中，计算机设备可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。例如，在一些实施例中，计算机设备可以仅包括存储器及处理器，在这样的实施例中，存储器及处理器的结构及功能与图10所示实施例一致，在此不再赘述。
应当理解，在本申请实施例中，处理器502可以是中央处理单元(Central Processing Unit，CPU)，该处理器502还可以是其他通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现成可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。其中，通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
在本申请的另一实施例中提供计算机可读存储介质。该计算机可读存储介质可以为易失性或非易失性的计算机可读存储介质，也可以是易失性的计算机可读存储介质。该计算机可读存储介质存储有计算机程序，当所述计算机程序被处理器执行时实现上述的基于多层关联知识图谱的信息预测方法。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，上述描述的设备、装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、计算机软件或者二者的结合来实现，为了清楚地说明硬件和软件的可互换性，在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请的范围。
在本申请所提供的几个实施例中，应该理解到，所揭露的设备、装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为逻辑功能划分，实际实现时可以有另外的划分方式，也可以将具有相同功能的单元集合成一个单元，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另外，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口、装置或单元的间接耦合或通信连接，也可以是电的，机械的或其它的形式连接。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本申请实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以是两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分，或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个计算机可读存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的计算机可读存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到各种等效的修改或替换，这些修改或替换都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应以权利要求的保护范围为准。

Claims (20)

  1. 一种基于多层关联知识图谱的信息预测方法，包括：
    根据预置的聚合规则对预存的历史数据信息中包含的历史数据进行聚合得到对应的多个数据点；
    根据预置的特征提取模型及每一所述数据点对应的历史数据，从每一所述数据点中提取得到对应的特征信息；
    根据每一所述数据点的特征信息对所述数据点进行分层得到多个数据层；
    根据所述历史数据之间的连接关系及每一所述数据层中数据点的特征信息生成多层关联知识图谱；
    若接收到用户输入的新增数据信息，根据所述特征提取模型获取与所述新增数据信息对应的新增数据特征信息；
    根据预置的判断条件及所述新增数据特征信息对所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点进行判断；
    若所述多层关联知识图谱中包含与所述新增数据信息相匹配的数据节点，获取所述多层关联知识图谱中与所述数据节点相匹配的索引连接信息作为对应的预测结果。
  2. 根据权利要求1所述的基于多层关联知识图谱的信息预测方法，其中，所述聚合规则包括关键词集合、相似度计算公式及相似度阈值，所述根据预置的聚合规则对预存的历史数据信息中包含的历史数据进行聚合得到对应的多个数据点，包括：
    对每一所述历史数据是否为文本信息进行判断；
    若所述历史数据为文本信息，从所述关键词集合中获取与历史数据相匹配的多个关键词；
    对包含相同关键词的多条所述历史数据进行聚合，得到对应的一个数据点；
    若所述历史数据不为文本信息，根据所述相似度计算公式对所述历史数据之间的相似度进行计算；
    对互相之间相似度大于所述相似度阈值的多条所述历史数据进行聚合，得到对应的一个数据点。
  3. 根据权利要求1所述的基于多层关联知识图谱的信息预测方法，其中，所述特征提取模型包括文本编码词典、特征提取神经网络及图像特征提取规则，所述根据预置的特征提取模型及每一所述数据点对应的历史数据，从每一所述数据点中提取得到对应的特征信息，包括：
    对所述数据点对应的历史数据是否均为文本信息进行判断；
    若所述数据点对应的历史数据均为文本信息，根据所述文本编码词典对每一所述数据点对应的多个关键词进行转换，得到每一所述数据点的编码信息；
    将所述编码信息输入所述特征提取神经网络进行计算，得到每一所述数据点的特征信息；
    若所述数据点对应的历史数据不均为文本信息，根据所述图像特征提取规则从所述数据点对应的历史数据中提取得到与所述数据点对应的特征信息。
  4. 根据权利要求3所述的基于多层关联知识图谱的信息预测方法，其中，所述图像特征提取规则包括对比度计算公式及溶解比例值，所述根据所述图像特征提取规则从所述数据点对应的历史数据中提取得到与所述数据点对应的特征信息，包括：
    对每一所述数据点对应的历史数据进行叠加得到每一数据点对应的叠加图像；
    根据所述对比度计算公式计算得到每一所述叠加图像的像素对比度信息；
    根据每一所述叠加图像的像素对比度信息及所述溶解比例值对每一所述叠加图像进行像素溶解得到图像轮廓信息；
    从每一所述叠加图像的图像轮廓信息中提取得到对应的轮廓尺寸信息及轮廓像素信息作为所述特征信息。
  5. 根据权利要求1所述的基于多层关联知识图谱的信息预测方法，其中，所述根据每一 所述数据点的特征信息对所述数据点进行分层得到多个数据层，包括：
    根据所述历史数据的类型及每一所述数据点对应的历史数据对每一所述数据点进行分类得到多个数据点集；
    根据预置的聚类规则及所述数据点的特征信息，对每一所述数据点集包含的数据点进行聚类得到对应的多个数据层。
  6. 根据权利要求1所述的基于多层关联知识图谱的信息预测方法，其中，所述根据所述历史数据之间的连接关系及每一所述数据层包含的历史数据的特征信息生成多层关联知识图谱，包括：
    生成与每一所述数据点对应的数据节点；
    分别对每一所述数据层及每一所述数据层包含的数据节点进行编码得到数据编码信息；
    根据所述历史数据的连接关系对所述数据节点及所述数据层进行倒排索引连接，得到对应的多层关联知识图谱。
  7. 根据权利要求6所述的基于多层关联知识图谱的信息预测方法，其中，所述判断条件包括匹配度计算公式及匹配度阈值，所述根据预置的判断条件及所述新增数据特征信息对所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点进行判断，包括：
    根据所述匹配度计算公式对所述新增数据特征信息与每一所述数据节点的特征信息之间的匹配度进行计算；
    判断匹配度大于所述匹配度阈值的数据节点的数量是否大于零，以得到所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点的判断结果。
  8. 一种基于多层关联知识图谱的信息预测装置，所述基于多层关联知识图谱的信息预测装置，包括：
    历史数据聚合单元，用于根据预置的聚合规则对预存的历史数据信息中包含的历史数据进行聚合得到对应的多个数据点；
    特征信息提取单元，用于根据预置的特征提取模型及每一所述数据点对应的历史数据，从每一所述数据点中提取得到对应的特征信息；
    数据层获取单元，用于根据每一所述数据点的特征信息对所述数据点进行分层得到多个数据层；
    知识图谱生成单元，用于根据所述历史数据之间的连接关系及每一所述数据层中数据点的特征信息生成多层关联知识图谱；
    新增数据特征信息获取单元，用于若接收到用户输入的新增数据信息，根据所述特征提取模型获取与所述新增数据信息对应的新增数据特征信息；
    判断单元，用于根据预置的判断条件及所述新增数据特征信息对所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点进行判断；
    预测结果获取单元，用于若所述多层关联知识图谱中包含与所述新增数据信息相匹配的数据节点，获取所述多层关联知识图谱中与所述数据节点相匹配的索引连接信息作为对应的预测结果。
  9. 一种计算机设备，其中，所述计算机设备包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序以实现以下步骤：
    根据预置的聚合规则对预存的历史数据信息中包含的历史数据进行聚合得到对应的多个数据点；
    根据预置的特征提取模型及每一所述数据点对应的历史数据，从每一所述数据点中提取得到对应的特征信息；
    根据每一所述数据点的特征信息对所述数据点进行分层得到多个数据层；
    根据所述历史数据之间的连接关系及每一所述数据层中数据点的特征信息生成多层关联 知识图谱；
    若接收到用户输入的新增数据信息，根据所述特征提取模型获取与所述新增数据信息对应的新增数据特征信息；
    根据预置的判断条件及所述新增数据特征信息对所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点进行判断；
    若所述多层关联知识图谱中包含与所述新增数据信息相匹配的数据节点，获取所述多层关联知识图谱中与所述数据节点相匹配的索引连接信息作为对应的预测结果。
  10. 根据权利要求9所述的计算机设备，其中，所述聚合规则包括关键词集合、相似度计算公式及相似度阈值，所述根据预置的聚合规则对预存的历史数据信息中包含的历史数据进行聚合得到对应的多个数据点，包括：
    对每一所述历史数据是否为文本信息进行判断；
    若所述历史数据为文本信息，从所述关键词集合中获取与历史数据相匹配的多个关键词；
    对包含相同关键词的多条所述历史数据进行聚合，得到对应的一个数据点；
    若所述历史数据不为文本信息，根据所述相似度计算公式对所述历史数据之间的相似度进行计算；
    对互相之间相似度大于所述相似度阈值的多条所述历史数据进行聚合，得到对应的一个数据点。
  11. 根据权利要求9所述的计算机设备，其中，所述特征提取模型包括文本编码词典、特征提取神经网络及图像特征提取规则，所述根据预置的特征提取模型及每一所述数据点对应的历史数据，从每一所述数据点中提取得到对应的特征信息，包括：
    对所述数据点对应的历史数据是否均为文本信息进行判断；
    若所述数据点对应的历史数据均为文本信息，根据所述文本编码词典对每一所述数据点对应的多个关键词进行转换，得到每一所述数据点的编码信息；
    将所述编码信息输入所述特征提取神经网络进行计算，得到每一所述数据点的特征信息；
    若所述数据点对应的历史数据不均为文本信息，根据所述图像特征提取规则从所述数据点对应的历史数据中提取得到与所述数据点对应的特征信息。
  12. 根据权利要求11所述的计算机设备，其中，所述图像特征提取规则包括对比度计算公式及溶解比例值，所述根据所述图像特征提取规则从所述数据点对应的历史数据中提取得到与所述数据点对应的特征信息，包括：
    对每一所述数据点对应的历史数据进行叠加得到每一数据点对应的叠加图像；
    根据所述对比度计算公式计算得到每一所述叠加图像的像素对比度信息；
    根据每一所述叠加图像的像素对比度信息及所述溶解比例值对每一所述叠加图像进行像素溶解得到图像轮廓信息；
    从每一所述叠加图像的图像轮廓信息中提取得到对应的轮廓尺寸信息及轮廓像素信息作为所述特征信息。
  13. 根据权利要求9所述的计算机设备，其中，所述根据每一所述数据点的特征信息对所述数据点进行分层得到多个数据层，包括：
    根据所述历史数据的类型及每一所述数据点对应的历史数据对每一所述数据点进行分类得到多个数据点集；
    根据预置的聚类规则及所述数据点的特征信息，对每一所述数据点集包含的数据点进行聚类得到对应的多个数据层。
  14. 根据权利要求9所述的计算机设备，其中，所述根据所述历史数据之间的连接关系及每一所述数据层包含的历史数据的特征信息生成多层关联知识图谱，包括：
    生成与每一所述数据点对应的数据节点；
    分别对每一所述数据层及每一所述数据层包含的数据节点进行编码得到数据编码信息；
    根据所述历史数据的连接关系对所述数据节点及所述数据层进行倒排索引连接，得到对应的多层关联知识图谱。
  15. 根据权利要求14所述的计算机设备，其中，所述判断条件包括匹配度计算公式及匹配度阈值，所述根据预置的判断条件及所述新增数据特征信息对所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点进行判断，包括：
    根据所述匹配度计算公式对所述新增数据特征信息与每一所述数据节点的特征信息之间的匹配度进行计算；
    判断匹配度大于所述匹配度阈值的数据节点的数量是否大于零，以得到所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点的判断结果。
  16. 一种计算机可读存储介质，其中，所述计算机可读存储介质存储有计算机程序，当所述计算机程序被处理器执行以下操作：
    根据预置的聚合规则对预存的历史数据信息中包含的历史数据进行聚合得到对应的多个数据点；
    根据预置的特征提取模型及每一所述数据点对应的历史数据，从每一所述数据点中提取得到对应的特征信息；
    根据每一所述数据点的特征信息对所述数据点进行分层得到多个数据层；
    根据所述历史数据之间的连接关系及每一所述数据层中数据点的特征信息生成多层关联知识图谱；
    若接收到用户输入的新增数据信息，根据所述特征提取模型获取与所述新增数据信息对应的新增数据特征信息；
    根据预置的判断条件及所述新增数据特征信息对所述多层关联知识图谱中是否包含与所述新增数据信息相匹配的数据节点进行判断；
    若所述多层关联知识图谱中包含与所述新增数据信息相匹配的数据节点，获取所述多层关联知识图谱中与所述数据节点相匹配的索引连接信息作为对应的预测结果。
  17. 根据权利要求16所述的计算机设备，其中，所述聚合规则包括关键词集合、相似度计算公式及相似度阈值，所述根据预置的聚合规则对预存的历史数据信息中包含的历史数据进行聚合得到对应的多个数据点，包括：
    对每一所述历史数据是否为文本信息进行判断；
    若所述历史数据为文本信息，从所述关键词集合中获取与历史数据相匹配的多个关键词；
    对包含相同关键词的多条所述历史数据进行聚合，得到对应的一个数据点；
    若所述历史数据不为文本信息，根据所述相似度计算公式对所述历史数据之间的相似度进行计算；
    对互相之间相似度大于所述相似度阈值的多条所述历史数据进行聚合，得到对应的一个数据点。
  18. 根据权利要求16所述的计算机设备，其中，所述特征提取模型包括文本编码词典、特征提取神经网络及图像特征提取规则，所述根据预置的特征提取模型及每一所述数据点对应的历史数据，从每一所述数据点中提取得到对应的特征信息，包括：
    对所述数据点对应的历史数据是否均为文本信息进行判断；
    若所述数据点对应的历史数据均为文本信息，根据所述文本编码词典对每一所述数据点对应的多个关键词进行转换，得到每一所述数据点的编码信息；
    将所述编码信息输入所述特征提取神经网络进行计算，得到每一所述数据点的特征信息；
    若所述数据点对应的历史数据不均为文本信息，根据所述图像特征提取规则从所述数据点对应的历史数据中提取得到与所述数据点对应的特征信息。
  19. 根据权利要求18所述的计算机设备，其中，所述图像特征提取规则包括对比度计算公式及溶解比例值，所述根据所述图像特征提取规则从所述数据点对应的历史数据中提取得到与所述数据点对应的特征信息，包括：
    对每一所述数据点对应的历史数据进行叠加得到每一数据点对应的叠加图像；
    根据所述对比度计算公式计算得到每一所述叠加图像的像素对比度信息；
    根据每一所述叠加图像的像素对比度信息及所述溶解比例值对每一所述叠加图像进行像素溶解得到图像轮廓信息；
    从每一所述叠加图像的图像轮廓信息中提取得到对应的轮廓尺寸信息及轮廓像素信息作为所述特征信息。
  20. 根据权利要求16所述的计算机设备，其中，所述根据每一所述数据点的特征信息对所述数据点进行分层得到多个数据层，包括：
    根据所述历史数据的类型及每一所述数据点对应的历史数据对每一所述数据点进行分类得到多个数据点集；
    根据预置的聚类规则及所述数据点的特征信息，对每一所述数据点集包含的数据点进行聚类得到对应的多个数据层。
PCT/CN2021/097105 2021-04-15 2021-05-31 基于多层关联知识图谱的信息预测方法、装置、设备及介质 WO2022217711A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110406359.9A CN112948508B (zh) 2021-04-15 2021-04-15 基于多层关联知识图谱的信息预测方法、装置、设备及介质
CN202110406359.9 2021-04-15

Publications (1)

Publication Number Publication Date
WO2022217711A1 true WO2022217711A1 (zh) 2022-10-20

Family

ID=76232627

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/097105 WO2022217711A1 (zh) 2021-04-15 2021-05-31 基于多层关联知识图谱的信息预测方法、装置、设备及介质

Country Status (2)

Country Link
CN (1) CN112948508B (zh)
WO (1) WO2022217711A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115860179A (zh) * 2022-11-01 2023-03-28 清华大学 轨迹预测方法、装置、设备、存储介质及程序产品

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114780864B (zh) * 2022-03-24 2023-01-31 天津大学 一种基于知识图谱的人-物-空间交互模型构建方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110321482A (zh) * 2019-06-11 2019-10-11 阿里巴巴集团控股有限公司 一种信息的推荐方法、装置及设备
CN111930774A (zh) * 2020-08-06 2020-11-13 全球能源互联网研究院有限公司 一种电力知识图谱本体自动构建方法及系统
CN112100369A (zh) * 2020-07-29 2020-12-18 浙江大学 结合语义的网络故障关联规则生成方法和网络故障检测方法
CN112163099A (zh) * 2020-09-24 2021-01-01 平安直通咨询有限公司上海分公司 基于知识图谱的文本识别方法、装置、存储介质和服务器
WO2021001047A1 (en) * 2019-07-04 2021-01-07 Siemens Aktiengesellschaft System, apparatus and method of managing knowledge generated from technical data

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8265778B2 (en) * 2010-06-17 2012-09-11 Microsoft Corporation Event prediction using hierarchical event features
CN107679110A (zh) * 2017-09-15 2018-02-09 广州唯品会研究院有限公司 结合文本分类与图片属性提取完善知识图谱的方法及装置
CN109697233B (zh) * 2018-12-03 2023-06-20 中电科大数据研究院有限公司 一种知识图谱体系搭建方法
CN111259149B (zh) * 2020-01-19 2024-03-12 清华大学 化学品事故分类方法、装置、计算机设备和存储介质
CN111694966B (zh) * 2020-06-10 2023-07-21 齐鲁工业大学 面向化工领域的多层次知识图谱构建方法及系统
CN111797222B (zh) * 2020-06-29 2023-12-22 平安国际智慧城市科技股份有限公司 课程知识图谱构建方法、装置、终端及存储介质
CN111831911B (zh) * 2020-07-16 2023-07-07 北京奇艺世纪科技有限公司 查询信息的处理方法、装置、存储介质和电子装置
CN112200317B (zh) * 2020-09-28 2024-05-07 西南电子技术研究所（中国电子科技集团公司第十研究所） 多模态知识图谱构建方法
CN112328741B (zh) * 2020-11-03 2022-02-18 平安科技（深圳）有限公司 基于人工智能的智能关联答复方法、装置、计算机设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110321482A (zh) * 2019-06-11 2019-10-11 阿里巴巴集团控股有限公司 一种信息的推荐方法、装置及设备
WO2021001047A1 (en) * 2019-07-04 2021-01-07 Siemens Aktiengesellschaft System, apparatus and method of managing knowledge generated from technical data
CN112100369A (zh) * 2020-07-29 2020-12-18 浙江大学 结合语义的网络故障关联规则生成方法和网络故障检测方法
CN111930774A (zh) * 2020-08-06 2020-11-13 全球能源互联网研究院有限公司 一种电力知识图谱本体自动构建方法及系统
CN112163099A (zh) * 2020-09-24 2021-01-01 平安直通咨询有限公司上海分公司 基于知识图谱的文本识别方法、装置、存储介质和服务器

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115860179A (zh) * 2022-11-01 2023-03-28 清华大学 轨迹预测方法、装置、设备、存储介质及程序产品

Also Published As

Publication number Publication date
CN112948508B (zh) 2024-04-30
CN112948508A (zh) 2021-06-11

Similar Documents

Publication Publication Date Title
Li et al. A principle component analysis-based random forest with the potential nearest neighbor method for automobile insurance fraud identification
Chino et al. Bowfire: detection of fire in still images by integrating pixel color and texture analysis
Fang et al. Android malware familial classification based on dex file section features
US11587234B2 (en) Generating class-agnostic object masks in digital images
WO2022121218A1 (zh) 智能图像识别方法、装置、计算机设备及存储介质
WO2022217711A1 (zh) 基于多层关联知识图谱的信息预测方法、装置、设备及介质
CN111325245B (zh) 重复图像识别方法、装置、电子设备及计算机可读存储介质
Shen et al. A direct formulation for totally-corrective multi-class boosting
US9734434B2 (en) Feature interpolation
CN111639607A (zh) 模型训练、图像识别方法和装置、电子设备及存储介质
Siddalingappa et al. Anomaly detection on medical images using autoencoder and convolutional neural network
Wang et al. A novel matrix of sequence descriptors for predicting protein-protein interactions from amino acid sequences
Chen et al. Reference-free method for investigating classification uncertainty in large-scale land cover datasets
Meng et al. Merged region based image retrieval
CN111444362A (zh) 恶意图片拦截方法、装置、设备和存储介质
CN113591881B (zh) 基于模型融合的意图识别方法、装置、电子设备及介质
CN116029760A (zh) 消息推送方法、装置、计算机设备和存储介质
CN115147633A (zh) 图像聚类方法、装置、设备及存储介质
WO2022267167A1 (zh) 文本类型智能识别方法、装置、设备及介质
Assiri et al. A hybrid ensemble method for accurate breast cancer tumor classification using state-of-the-art classification learning algorithms
CN113343102A (zh) 基于特征筛选的数据推荐方法、装置、电子设备及介质
Bar et al. Facilitating high‐dimensional transparent classification via empirical Bayes variable selection
Khan et al. Towards fairness in multimodal scene graph generation: Mitigating biases in datasets, knowledge sources and models
JP2020140488A (ja) 情報処理装置、情報処理方法及びプログラム
KR102569976B1 (ko) 의료 영상 처리 방법

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21936583

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21936583

Country of ref document: EP

Kind code of ref document: A1