A method and device for identifying, classifying and structurally analyzing text attribute features

Abstract

The present invention discloses a method and device for identifying, classifying and structurally analyzing text attribute features. The method for identifying text attribute features includes: generating a grammatical structure according to sentences in a target text; generating a data structure according to node relationships in the grammatical structure; generating a first input vector according to the data structure; determining the probability that each sentence contains an attribute feature text according to the first input vector and a preset text attribute feature classification model; and identifying the text attribute features in the target text according to the probability. By implementing the present invention, the text attribute features in the target text can be accurately identified, the meaning of the text content can be identified, and the deeper meaning of the text can be excavated, which can not only enrich the content of text recognition, but also provide more comprehensive data and content support for subsequent analysis and processing based on the text recognition content.

Description

Technical Field

[0001] The present invention relates to the technical field of data mining, and in particular to a method and device for identifying, classifying and structurally analyzing text attribute features. Background Art

[0002] With the development of intellectual property, the huge value of patent data has gradually attracted people's attention. How to effectively mine the value of this data is very important. However, the existing text recognition and analysis methods only stay at the level of identifying and analyzing the text part of speech, thereby analyzing the basic structure of the text, and cannot know the deeper meaning of the text. When users want to search based on a class of words that represent certain characteristics of the text (for example, words that express effects, words that express praise or derogatory meanings, etc.) as keywords, the existing text recognition and analysis methods cannot effectively identify such words. Therefore, the user's search needs are still not met. It can be seen that based on the existing text recognition and analysis methods, the development and utilization of text content is still subject to great restrictions. Therefore, there is an urgent need for a technology that can deeply mine text content. Summary of the invention

[0003] In view of this, an embodiment of the present invention provides a method and device for identifying, classifying and structurally analyzing text attribute features to solve the problem that the development and utilization of text content based on existing text recognition and analysis methods is still subject to significant limitations.

[0004] According to the first aspect, an embodiment of the present invention provides a method for identifying text attribute features, including: generating a grammatical structure based on sentences in a target text; generating a data structure based on node relationships in the grammatical structure; generating a first input vector based on the data structure; determining the probability that each of the sentences contains attribute feature text based on the first input vector and a preset text attribute feature classification model; and identifying the text attribute features in the target text based on the probability.

[0005] In combination with the first aspect, in the first implementation of the first aspect, the generation of a grammatical structure based on sentences in the target text includes: respectively identifying the words of each sentence in the target text and constructing a word directed graph; calculating the shortest path from the first node to the last node in the word directed graph as the word segmentation result of each sentence; constructing a word sequence according to the word segmentation result of each sentence; generating an input vector according to adjacent words in the word sequence; obtaining an output vector according to a preset neural network model and the input vector; calculating the cosine value of the angle between the input vector and the output vector; constructing a combined node for two adjacent words with the largest cosine value of the angle until the root node of the word sequence is generated; and determining the grammatical structure of the word sequence according to the combined node and the root node.

[0006] In combination with the first aspect, in the second implementation of the first aspect, the preset text attribute feature classification model is an efficacy sentence classification model, and the efficacy sentence classification model is constructed through the following process: obtaining efficacy sentence samples, the efficacy sentence samples including preset efficacy tags and sample sentences; generating a first grammatical structure according to the efficacy sentence samples; generating a first feature list according to the node relationship in the first grammatical structure; generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy tags and sample sentences; training the preset classification model according to the first classification input vector and the first classification output vector to generate the efficacy sentence classification model.

[0007] In combination with the first aspect, in the third implementation of the first aspect, the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process: obtaining a functional phrase sample, the functional phrase sample including a preset functional tag and a sample phrase; generating a second grammatical structure according to the functional phrase sample; generating a second feature list according to the node relationship in the second grammatical structure; generating a second data structure according to the preset functional phrase; generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset functional tag and the sample phrase; training the preset classification model according to the second classification input vector and the second classification output vector to generate the preset functional phrase classification model.

[0008] In combination with the first aspect, in a fourth implementation of the first aspect, the probability that each of the sentences contains attribute feature text is determined based on the first input vector and a preset text attribute feature classification model, including at least one of the following steps: determining a first probability that each of the sentences contains the attribute feature based on the first input vector and a preset text attribute feature classification model; determining a second probability that a phrase in each of the sentences contains the attribute feature based on the first input vector and a preset text attribute feature classification model; determining a third probability that a paragraph in each of the sentences contains the attribute feature based on the first input vector and a preset text attribute feature classification model.

[0009] According to a second aspect, an embodiment of the present invention provides a method for classifying text attribute features, comprising: identifying text attribute features in a target text according to sentences in the target text by using a method for identifying text attribute features as described in the first aspect or any one of the embodiments of the first aspect; constructing a text attribute feature set according to the recognition result of the target text; constructing a structure tree according to the text attribute feature set; respectively determining the class center point of each leaf node in the structure tree, and forming a class center set with multiple class center points; generating a class center vector according to the multiple class center points, and generating a leaf node vector according to the leaf nodes of the structure tree; and determining the classification node of each leaf node according to the class center vector and the leaf node vector classification node.

[0010] In combination with the second aspect, in the first implementation of the second aspect, respectively determining the class center point of each leaf node in the structure tree includes: respectively taking one of the leaf nodes in the structure tree as the center node, calculating the average distance from the leaf nodes in the structure tree to the center node; and determining the leaf node with the farthest average distance as the class center point.

[0011] In combination with the second aspect, in the second implementation of the second aspect, the classification node of each leaf node is determined based on the class center vector and the leaf node vector classification node, including: calculating the Euclidean distance from the leaf node vector to the class center vector; for each leaf node, taking the class center point corresponding to the class center vector with the smallest Euclidean distance as the classification node of each leaf node.

[0012] According to the third aspect, an embodiment of the present invention provides a method for constructing a classification model of text attribute features, including: obtaining an engineering parameter corpus sample and a text attribute feature sample, the engineering parameter corpus sample including an engineering parameter text attribute feature and a text attribute feature type tag, the text attribute feature sample being identified and generated according to the text attribute feature identification method described in the first aspect or any implementation scheme of the first aspect; generating an engineering parameter input vector according to the text attribute feature sample, and generating an engineering parameter output vector according to the engineering parameter corpus sample; training an SVM model according to the engineering parameter input vector and the engineering parameter output vector to construct an engineering parameter classification model.

[0013] According to the fourth aspect, an embodiment of the present invention provides a method for classifying text attribute features, which is characterized in that it includes: obtaining a target text, identifying the text attribute features of the target text by using the text attribute feature recognition method as described in the first aspect or any implementation method of the first aspect; and generating an engineering parameter classification result corresponding to the text attribute features based on the text attribute features and a preset engineering parameter classification model.

[0014] In combination with the fourth aspect, in a first implementation of the fourth aspect, the engineering parameter classification model is constructed according to the method for constructing a classification model of text attribute features described in the third aspect.

[0015] According to the fifth aspect, an embodiment of the present invention provides a text structure analysis method, including: identifying phrases in each paragraph of the target text, and forming phrase sets according to each paragraph; performing structured analysis on each phrase set to generate a structured set; identifying text attribute features in the paragraphs of the structured set according to the text attribute feature identification method described in the first aspect or any implementation method of the first aspect; and establishing an association relationship between the structured set and the phrase set based on the identified text attribute features.

[0016] In combination with the fifth aspect, in the first implementation of the fifth aspect, if the paragraph of the structured set does not contain text attribute features, the first phrase set that is closest to the current phrase and located after the current phrase is selected to establish an association relationship with the structured set.

[0017] In combination with the first implementation of the fifth aspect, in the second implementation of the fifth aspect, if the first phrase set is not found, a second phrase set that is closest to the current phrase and located in front of the current phrase is selected to establish an association relationship with the structured set.

[0018] In combination with the second implementation of the fifth aspect, in the third implementation of the fifth aspect, if neither the first phrase set nor the second phrase set exists, the current phrase is ignored.

[0019] According to the sixth aspect, an embodiment of the present invention provides a device for identifying text attribute features, including: a grammatical structure generation module, used to generate a grammatical structure according to sentences in a target text; a data structure generation module, used to generate a data structure according to the node relationship in the grammatical structure; a first input vector generation module, used to generate a first input vector according to the data structure; a probability determination module, used to determine the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model; and a text recognition module, used to recognize the text attribute features in the target text according to the probability.

[0020] According to a seventh aspect, an embodiment of the present invention provides a classification device for text attribute features, comprising: a text recognition module, used to recognize text attribute features in the target text according to sentences in the target text by using the recognition method for text attribute features as described in the first aspect or any one of the embodiments of the first aspect; a text attribute feature set construction module, used to construct a text attribute feature set according to the recognition result of the target text; a structure tree construction module, used to construct a structure tree according to the text attribute feature set; a class center set construction module, used to respectively determine the class center point of each leaf node in the structure tree, and form a class center set with multiple class center points; a vector generation module, used to generate a class center vector according to the multiple class center points, and generate a leaf node vector according to the leaf nodes of the structure tree; a classification node determination module, used to determine the classification node of each leaf node according to the class center vector and the leaf node vector classification node.

[0021] According to the eighth aspect, an embodiment of the present invention provides a device for constructing a classification model of text attribute features, including: a sample acquisition module, used to obtain engineering parameter corpus samples and text attribute feature samples, the engineering parameter corpus samples include engineering parameter text attribute features and text attribute feature type tags, and the text attribute feature samples are identified and generated according to the text attribute feature recognition method described in the first aspect or any implementation method of the first aspect; a vector generation module, used to generate an engineering parameter input vector according to the text attribute feature sample, and generate an engineering parameter output vector according to the engineering parameter corpus sample; a model construction module, used to train an SVM model according to the engineering parameter input vector and the engineering parameter output vector to construct an engineering parameter classification model.

[0022] According to the ninth aspect, an embodiment of the present invention provides a classification device for text attribute features, including: a text recognition module, used to obtain a target text, and identify the efficacy phrases of the target text by using the text attribute feature recognition method as described in the first aspect or any embodiment of the first aspect; a text classification module, used to generate an engineering parameter classification result corresponding to the efficacy phrase based on the text attribute features and a preset engineering parameter classification model.

[0023] According to the tenth aspect, an embodiment of the present invention provides a text structure analysis device, including: a phrase set generation module, used to identify phrases in each paragraph of the target text, and form phrase sets according to each paragraph; a structured set generation module, used to perform structured analysis on each phrase set to generate a structured set; a text recognition module, used to identify text attribute features in the paragraphs of the structured set according to the text attribute feature recognition method described in the first aspect or any implementation method of the first aspect; a text relationship construction module, used to establish an association relationship between the structured set and the phrase set according to the recognized text attribute features.

[0024] According to the eleventh aspect, an embodiment of the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to execute the text attribute feature recognition method described in the first aspect or any one of the first embodiments, or executes the text attribute feature classification method described in the second aspect or any one of the second embodiments, or executes the text attribute feature construction method described in the third aspect, or executes the text attribute feature classification method described in the fourth aspect or any one of the fourth embodiments, or executes the text structure analysis method described in the fifth aspect or any one of the fifth embodiments.

[0025] According to the twelfth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the text attribute feature recognition method described in the first aspect or any one of the embodiments of the first aspect, or execute the text attribute feature classification method described in the second aspect or any one of the embodiments of the second aspect, or execute the text attribute feature classification model construction method described in the third aspect, or execute the text attribute feature classification method described in the fourth aspect or any one of the embodiments of the fourth aspect, or execute the text structure analysis method described in the fifth aspect or any one of the embodiments of the fifth aspect.

[0026] Solution 1. A method for identifying text attribute features, comprising:

[0027] Get the data structure in the target text;

[0028] generating a first input vector according to the data structure;

[0029] Determining the probability that each of the target texts contains attribute feature text according to the first input vector and a preset text attribute feature classification model;

[0030] The text attribute features in the target text are identified according to the probability.

[0031] Solution 2. According to the method for identifying text attribute features described in Solution 1, the step of obtaining the data structure of the target text comprises:

[0032] Generate grammatical structures based on sentences in the target text;

[0033] A data structure is generated according to the node relationship in the grammatical structure.

[0034] Solution 3. According to the method for identifying text attribute features described in Solution 2, the step of generating a grammatical structure based on sentences in the target text comprises:

[0035] Respectively identifying words of each sentence in the target text and constructing a word directed graph;

[0036] Calculate the shortest path from the first node to the last node in the word directed graph as the word segmentation result of each of the sentences;

[0037] Construct a word sequence based on the word segmentation results of each sentence;

[0038] Generating an input vector based on adjacent words in the word sequence;

[0039] Obtaining an output vector according to a preset neural network model and the input vector;

[0040] Calculating the cosine value of the angle between the input vector and the output vector;

[0041] Two adjacent words with the largest angle cosine value are used to construct a combined node until the root node of the word sequence is generated;

[0042] The grammatical structure of the word sequence is determined according to the combination node and the root node.

[0043] Solution 4. According to the text attribute feature recognition method described in Solution 3, the preset text attribute feature classification model is an efficacy sentence classification model, and the efficacy sentence classification model is constructed by the following process:

[0044] Acquire a sample of efficacy statement, wherein the sample of efficacy statement includes a preset efficacy mark and a sample statement;

[0045] Generate a first grammatical structure according to the efficacy sentence sample;

[0046] generating a first feature list according to the node relationship in the first grammatical structure;

[0047] Generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy mark and the sample sentence;

[0048] The preset classification model is trained according to the first classification input vector and the first classification output vector to generate the efficacy statement classification model.

[0049] Solution 5. According to the text attribute feature recognition method described in Solution 4, the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process:

[0050] Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase;

[0051] generating a second grammatical structure according to the functional phrase sample;

[0052] generating a second feature list according to the node relationship in the second grammatical structure;

[0053] Generate a second data structure according to a preset efficacy phrase;

[0054] generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase;

[0055] The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

[0056] Solution 6. According to the text attribute feature recognition method described in Solution 5, the probability of each of the sentences containing attribute feature text is determined based on the first input vector and a preset text attribute feature classification model, and at least one of the following steps is included:

[0057] Determining a first probability that each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0058] Determining a second probability that a phrase in each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0059] A third probability that a paragraph in each of the sentences contains an attribute feature is determined based on the first input vector and a preset text attribute feature classification model.

[0060] Solution 7. According to the text attribute feature recognition method described in Solution 4, the probability of each of the sentences containing attribute feature text is determined based on the first input vector and a preset text attribute feature classification model, and at least one of the following steps is included:

[0061] Determining a first probability that each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0062] Determining a second probability that a phrase in each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0063] A third probability that a paragraph in each of the sentences contains an attribute feature is determined based on the first input vector and a preset text attribute feature classification model.

[0064] Solution 8. According to the text attribute feature recognition method described in Solution 3, the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process:

[0065] Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase;

[0066] generating a second grammatical structure according to the functional phrase sample;

[0067] generating a second feature list according to the node relationship in the second grammatical structure;

[0068] Generate a second data structure according to a preset efficacy phrase;

[0069] generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase;

[0070] The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

[0071] Solution 9. According to the text attribute feature recognition method described in Solution 8, the probability of each of the sentences containing attribute feature text is determined based on the first input vector and a preset text attribute feature classification model, and at least one of the following steps is included:

[0072] Determining a first probability that each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0073] Determining a second probability that a phrase in each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0074] A third probability that a paragraph in each of the sentences contains an attribute feature is determined based on the first input vector and a preset text attribute feature classification model.

[0075] Solution 10. According to the text attribute feature recognition method described in Solution 1, the preset text attribute feature classification model is an efficacy sentence classification model, and the efficacy sentence classification model is constructed by the following process:

[0076] Acquire a sample of efficacy statement, wherein the sample of efficacy statement includes a preset efficacy mark and a sample statement;

[0077] Generate a first grammatical structure according to the efficacy sentence sample;

[0078] generating a first feature list according to the node relationship in the first grammatical structure;

[0079] Generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy mark and the sample sentence;

[0080] The preset classification model is trained according to the first classification input vector and the first classification output vector to generate the efficacy statement classification model.

[0081] Solution 11. According to the text attribute feature recognition method of Solution 10, the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process:

[0082] Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase;

[0083] generating a second grammatical structure according to the functional phrase sample;

[0084] generating a second feature list according to the node relationship in the second grammatical structure;

[0085] Generate a second data structure according to a preset efficacy phrase;

[0086] generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase;

[0087] The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

[0088] Solution 12. According to the text attribute feature recognition method of Solution 11, the probability of each of the sentences containing attribute feature text is determined according to the first input vector and a preset text attribute feature classification model, and at least one of the following steps is included:

[0089] Determining a first probability that each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0090] Determining a second probability that a phrase in each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0091] A third probability that a paragraph in each of the sentences contains an attribute feature is determined based on the first input vector and a preset text attribute feature classification model.

[0092] Solution 13. According to the text attribute feature recognition method described in Solution 1, the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process:

[0093] Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase;

[0094] generating a second grammatical structure according to the functional phrase sample;

[0095] generating a second feature list according to the node relationship in the second grammatical structure;

[0096] generating a second data structure according to a preset efficacy phrase;

[0097] generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase;

[0098] The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

[0099] Solution 14. According to the text attribute feature recognition method described in Solution 13, the probability of each target text containing an attribute feature text is determined based on the first input vector and a preset text attribute feature classification model, and at least one of the following steps is included:

[0100] Determining a first probability that each of the target texts contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0101] Determining a second probability that each phrase in the target text contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0102] A third probability that each paragraph in the target text contains an attribute feature is determined according to the first input vector and a preset text attribute feature classification model.

[0103] Solution 15. According to the text attribute feature recognition method described in Solution 1, the probability of each target text containing an attribute feature text is determined based on the first input vector and a preset text attribute feature classification model, and at least one of the following steps is included:

[0104] Determining a first probability that each of the target texts contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0105] Determining a second probability that each phrase in the target text contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0106] A third probability that each paragraph in the target text contains an attribute feature is determined according to the first input vector and a preset text attribute feature classification model.

[0107] Solution 16. A method for identifying text attribute features, comprising:

[0108] Generate grammatical structures based on sentences in the target text;

[0109] Generate a data structure according to the node relationship in the grammatical structure;

[0110] generating a first input vector according to the data structure;

[0111] Determining the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model;

[0112] Identify text attribute features in the target text according to the probability;

[0113] The step of generating a grammatical structure according to sentences in the target text includes:

[0114] Respectively identifying words of each sentence in the target text and constructing a word directed graph;

[0115] Calculate the shortest path from the first node to the last node in the word directed graph as the word segmentation result of each of the sentences;

[0116] Construct a word sequence based on the word segmentation results of each sentence;

[0117] Generating an input vector based on adjacent words in the word sequence;

[0118] Obtaining an output vector according to a preset neural network model and the input vector;

[0119] Calculating the cosine value of the angle between the input vector and the output vector;

[0120] Two adjacent words with the largest angle cosine value are used to construct a combined node until the root node of the word sequence is generated;

[0121] The grammatical structure of the word sequence is determined according to the combination node and the root node.

[0122] Solution 17. According to the text attribute feature recognition method of Solution 16, the preset text attribute feature classification model is an efficacy sentence classification model, and the efficacy sentence classification model is constructed by the following process:

[0123] Acquire a sample of efficacy statement, wherein the sample of efficacy statement includes a preset efficacy mark and a sample statement;

[0124] Generate a first grammatical structure according to the efficacy sentence sample;

[0125] generating a first feature list according to the node relationship in the first grammatical structure;

[0126] Generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy mark and the sample sentence;

[0127] The preset classification model is trained according to the first classification input vector and the first classification output vector to generate the efficacy statement classification model.

[0128] Solution 18. According to the text attribute feature recognition method of Solution 17, the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process:

[0129] Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase;

[0130] generating a second grammatical structure according to the functional phrase sample;

[0131] generating a second feature list according to the node relationship in the second grammatical structure;

[0132] Generate a second data structure according to a preset efficacy phrase;

[0133] generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase;

[0134] The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

[0135] Solution 19. According to the text attribute feature recognition method of Solution 16, the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process:

[0136] Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase;

[0137] generating a second grammatical structure according to the functional phrase sample;

[0138] generating a second feature list according to the node relationship in the second grammatical structure;

[0139] generating a second data structure according to a preset efficacy phrase;

[0140] generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase;

[0141] The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

[0142] Solution 20. A method for identifying text attribute features, comprising:

[0143] Generate grammatical structures based on sentences in the target text;

[0144] Generate a data structure according to the node relationship in the grammatical structure;

[0145] generating a first input vector according to the data structure;

[0146] Determining the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model;

[0147] Identify text attribute features in the target text according to the probability;

[0148] The preset text attribute feature classification model is an efficacy sentence classification model, which is constructed by the following process:

[0149] Acquire a sample of efficacy statement, wherein the sample of efficacy statement includes a preset efficacy mark and a sample statement;

[0150] Generate a first grammatical structure according to the efficacy sentence sample;

[0151] generating a first feature list according to the node relationship in the first grammatical structure;

[0152] Generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy mark and the sample sentence;

[0153] The preset classification model is trained according to the first classification input vector and the first classification output vector to generate the efficacy statement classification model.

[0154] Solution 21. According to the text attribute feature recognition method of Solution 20, the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process:

[0155] Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase;

[0156] generating a second grammatical structure according to the functional phrase sample;

[0157] generating a second feature list according to the node relationship in the second grammatical structure;

[0158] generating a second data structure according to a preset efficacy phrase;

[0159] generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase;

[0160] The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

[0161] Solution 22. A method for identifying text attribute features, comprising:

[0162] Generate grammatical structures based on sentences in the target text;

[0163] Generate a data structure according to the node relationship in the grammatical structure;

[0164] generating a first input vector according to the data structure;

[0165] Determining the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model;

[0166] Identify text attribute features in the target text according to the probability;

[0167] The preset text attribute feature classification model is a functional phrase classification model, which is constructed by the following process:

[0168] Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase;

[0169] generating a second grammatical structure according to the functional phrase sample;

[0170] generating a second feature list according to the node relationship in the second grammatical structure;

[0171] generating a second data structure according to a preset efficacy phrase;

[0172] generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase;

[0173] The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

[0174] Solution 23. The method for recognizing text attribute features according to Solution 17, 18, 19 or 21 is characterized in that determining the probability that each of the sentences contains attribute feature text based on the first input vector and a preset text attribute feature classification model comprises at least one of the following steps:

[0175] Determining a first probability that each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0176] Determining a second probability that a phrase in each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0177] A third probability that a paragraph in each of the sentences contains an attribute feature is determined based on the first input vector and a preset text attribute feature classification model.

[0178] Scheme 24. A method for classifying text attribute features, comprising:

[0179] Identifying text attribute features in the target text according to sentences in the target text by using the text attribute feature identification method described in any one of schemes 1-23;

[0180] Constructing a text attribute feature set according to the recognition result of the target text;

[0181] Constructing a structure tree according to the text attribute feature set;

[0182] Determine the class center point of each leaf node in the structure tree respectively, and form a class center set with multiple class center points;

[0183] Generate a class center vector according to the plurality of class center points, and generate a leaf node vector according to the leaf nodes of the structure tree;

[0184] The classification node of each leaf node is determined according to the class center vector and the leaf node vector classification node.

[0185] Solution 25. According to the text attribute feature classification method of Solution 24, the step of respectively determining the class center point of each leaf node in the structure tree comprises:

[0186] Taking one of the leaf nodes in the structure tree as the central node respectively, and calculating the average distance from the leaf node in the structure tree to the central node;

[0187] The leaf node with the farthest average distance is determined as the cluster center point.

[0188] Solution 26. According to the text attribute feature classification method of Solution 24, the step of determining the classification node of each leaf node according to the class center vector and the leaf node vector classification node includes:

[0189] Calculating the Euclidean distance from the leaf node vector to the class center vector;

[0190] For each leaf node, the class center point corresponding to the class center vector with the smallest Euclidean distance is used as the classification node of each leaf node.

[0191] Solution 27. A method for constructing a classification model of text attribute features, comprising:

[0192] Acquire an engineering parameter corpus sample and a text attribute feature sample, wherein the engineering parameter corpus sample includes an engineering parameter text attribute feature and a text attribute feature type tag, and the text attribute feature sample is generated by identifying the text attribute feature according to any one of schemes 1-23;

[0193] Generating an engineering parameter input vector according to the text attribute feature sample, and generating an engineering parameter output vector according to the engineering parameter corpus sample;

[0194] The SVM model is trained according to the engineering parameter input vector and the engineering parameter output vector to construct an engineering parameter classification model.

[0195] Scheme 28. A method for classifying text attribute features, comprising:

[0196] Acquire a target text, and identify text attribute features of the target text by using a text attribute feature recognition method as described in any one of Schemes 1-23;

[0197] An engineering parameter classification result corresponding to the text attribute feature is generated according to the text attribute feature and a preset engineering parameter classification model.

[0198] Scheme 29. According to the classification method of text attribute features described in Scheme 28, the engineering parameter classification model is constructed according to the construction method of the classification model of text attribute features described in Scheme 27.

[0199] Scheme 30. A text structure analysis method, comprising:

[0200] Identify phrases in each paragraph of the target text and form phrase sets based on each paragraph;

[0201] Perform structural analysis on each phrase set to generate a structured set;

[0202] Identify text attribute features in paragraphs of the structured set according to the text attribute feature identification method described in any one of Schemes 1-23;

[0203] An association relationship between the structured set and the phrase set is established according to the recognized text attribute features.

[0204] Solution 31. The text structure analysis method according to Solution 30,

[0205] If the paragraph of the structured set does not contain the text attribute feature, the first phrase set which is closest to the current phrase and located after the current phrase is selected and an association relationship is established with the structured set.

[0206] Solution 32. The text structure analysis method according to Solution 31,

[0207] If the first phrase set is not found, a second phrase set which is closest to the current phrase and located before the current phrase is selected and an association relationship is established with the structured set.

[0208] Solution 33. The text structure analysis method according to Solution 32,

[0209] If neither the first phrase set nor the second phrase set exists, the current phrase is ignored.

[0210] Solution 34. A device for identifying text attribute features, comprising:

[0211] A data structure acquisition module is used to acquire the data structure of the target text;

[0212] A first input vector generating module, used for generating a first input vector according to the data structure;

[0213] A probability determination module, used to determine the probability that each of the target texts contains attribute feature text according to the first input vector and a preset text attribute feature classification model;

[0214] A text recognition module, used for recognizing text attribute features in the target text according to the probability;

[0215] The preset text attribute feature classification model is an efficacy sentence classification model, which is constructed by the following process:

[0216] Acquire a sample of efficacy statement, wherein the sample of efficacy statement includes a preset efficacy mark and a sample statement;

[0217] Generate a first grammatical structure according to the efficacy sentence sample;

[0218] generating a first feature list according to the node relationship in the first grammatical structure;

[0219] Generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy mark and the sample sentence;

[0220] The preset classification model is trained according to the first classification input vector and the first classification output vector to generate the efficacy statement classification model.

[0221] Solution 35. A device for identifying text attribute features, comprising:

[0222] A data structure acquisition module is used to acquire the data structure of the target text;

[0223] A first input vector generating module, used for generating a first input vector according to the data structure;

[0224] A probability determination module, used to determine the probability that the target text contains attribute feature text according to the first input vector and a preset text attribute feature classification model;

[0225] A text recognition module, used for recognizing text attribute features in the target text according to the probability;

[0226] The preset text attribute feature classification model is a functional phrase classification model, which is constructed by the following process:

[0227] Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase;

[0228] generating a second grammatical structure according to the functional phrase sample;

[0229] generating a second feature list according to the node relationship in the second grammatical structure;

[0230] generating a second data structure according to a preset efficacy phrase;

[0231] generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase;

[0232] The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

[0233] Solution 36. A device for identifying text attribute features, comprising:

[0234] A data structure acquisition module is used to acquire the data structure of the target text;

[0235] A first input vector generating module, used for generating a first input vector according to the data structure;

[0236] A probability determination module, used to determine the probability that the target text contains attribute feature text according to the first input vector and a preset text attribute feature classification model;

[0237] A text recognition module, used for recognizing text attribute features in the target text according to the probability;

[0238] Wherein, the data structure acquisition module is specifically used for:

[0239] Generate grammatical structures based on sentences in the target text;

[0240] Generate a data structure according to the node relationship in the grammatical structure;

[0241] The step of generating a grammatical structure according to sentences in the target text includes:

[0242] Respectively identifying words of each sentence in the target text and constructing a word directed graph;

[0243] Calculate the shortest path from the first node to the last node in the word directed graph as the word segmentation result of each of the sentences;

[0244] Construct a word sequence based on the word segmentation results of each sentence;

[0245] Generating an input vector based on adjacent words in the word sequence;

[0246] Obtaining an output vector according to a preset neural network model and the input vector;

[0247] Calculating the cosine value of the angle between the input vector and the output vector;

[0248] Two adjacent words with the largest angle cosine value are used to construct a combined node until the root node of the word sequence is generated;

[0249] The grammatical structure of the word sequence is determined according to the combination node and the root node.

[0250] Solution 37. A device for identifying text attribute features, comprising:

[0251] A grammatical structure generation module, used for generating a grammatical structure according to sentences in a target text;

[0252] A data structure generation module, used to generate a data structure according to the node relationship in the grammatical structure;

[0253] A first input vector generating module, used for generating a first input vector according to the data structure;

[0254] A probability determination module, used to determine the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model;

[0255] A text recognition module, used for recognizing text attribute features in the target text according to the probability;

[0256] Wherein, the grammatical structure generation module is specifically used for:

[0257] Respectively identifying words of each sentence in the target text and constructing a word directed graph;

[0258] Calculate the shortest path from the first node to the last node in the word directed graph as the word segmentation result of each of the sentences;

[0259] Construct a word sequence based on the word segmentation results of each sentence;

[0260] Generating an input vector based on adjacent words in the word sequence;

[0261] Obtaining an output vector according to a preset neural network model and the input vector;

[0262] Calculating the cosine value of the angle between the input vector and the output vector;

[0263] Two adjacent words with the largest angle cosine value are used to construct a combined node until the root node of the word sequence is generated;

[0264] The grammatical structure of the word sequence is determined according to the combination node and the root node.

[0265] Solution 38. A device for identifying text attribute features, comprising:

[0266] A grammatical structure generation module, used for generating a grammatical structure according to sentences in a target text;

[0267] A data structure generation module, used to generate a data structure according to the node relationship in the grammatical structure;

[0268] A first input vector generating module, used for generating a first input vector according to the data structure;

[0269] A probability determination module, used to determine the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model;

[0270] A text recognition module, used for recognizing text attribute features in the target text according to the probability;

[0271] The preset text attribute feature classification model is an efficacy sentence classification model, which is constructed by the following process:

[0272] Acquire a sample of efficacy statement, wherein the sample of efficacy statement includes a preset efficacy mark and a sample statement;

[0273] Generate a first grammatical structure according to the efficacy sentence sample;

[0274] generating a first feature list according to the node relationship in the first grammatical structure;

[0275] Generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy mark and the sample sentence;

[0276] The preset classification model is trained according to the first classification input vector and the first classification output vector to generate the efficacy statement classification model.

[0277] Solution 39. A device for identifying text attribute features, comprising:

[0278] A grammatical structure generation module, used for generating a grammatical structure according to sentences in a target text;

[0279] A data structure generation module, used to generate a data structure according to the node relationship in the grammatical structure;

[0280] A first input vector generating module, used for generating a first input vector according to the data structure;

[0281] A probability determination module, used to determine the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model;

[0282] A text recognition module, used for recognizing text attribute features in the target text according to the probability;

[0283] The preset text attribute feature classification model is a functional phrase classification model, which is constructed by the following process:

[0284] Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase;

[0285] generating a second grammatical structure according to the functional phrase sample;

[0286] generating a second feature list according to the node relationship in the second grammatical structure;

[0287] generating a second data structure according to a preset efficacy phrase;

[0288] generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase;

[0289] The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

[0290] Solution 40. A text attribute feature classification device, comprising:

[0291] A text recognition module, configured to recognize text attribute features in a target text according to sentences in the target text by using a text attribute feature recognition method as described in any one of schemes 1 to 23;

[0292] A text attribute feature set construction module, used to construct a text attribute feature set according to the recognition result of the target text;

[0293] A structure tree construction module, used to construct a structure tree according to the text attribute feature set;

[0294] A class center set construction module is used to determine the class center point of each leaf node in the structure tree respectively, and form a class center set with multiple class center points;

[0295] A vector generation module, used to generate a class center vector according to the multiple class center points, and to generate a leaf node vector according to the leaf nodes of the structure tree;

[0296] The classification node determination module is used to determine the classification node of each leaf node according to the class center vector and the leaf node vector classification node.

[0297] Solution 41. A device for constructing a classification model of text attribute features, comprising:

[0298] A sample acquisition module, used to acquire an engineering parameter corpus sample and a text attribute feature sample, wherein the engineering parameter corpus sample includes an engineering parameter text attribute feature and a text attribute feature type tag, and the text attribute feature sample is generated by identifying the text attribute feature according to any one of schemes 1-23;

[0299] A vector generation module, used to generate an engineering parameter input vector according to the text attribute feature sample, and to generate an engineering parameter output vector according to the engineering parameter corpus sample;

[0300] The model building module is used to train the SVM model according to the engineering parameter input vector and the engineering parameter output vector to build an engineering parameter classification model.

[0301] Solution 42. A text attribute feature classification device, comprising:

[0302] A text recognition module, used to obtain a target text, and recognize the efficacy phrases of the target text by using the text attribute feature recognition method described in any one of Schemes 1-23;

[0303] The text classification module is used to generate the engineering parameter classification result corresponding to the efficacy phrase according to the text attribute characteristics and a preset engineering parameter classification model.

[0304] Solution 43. A text structure analysis device, comprising:

[0305] A phrase set generation module is used to identify phrases in each paragraph of the target text and form phrase sets according to each paragraph;

[0306] A structured set generation module is used to perform structural analysis on each phrase set to generate a structured set;

[0307] A text recognition module, used to recognize text attribute features in the paragraphs of the structured set according to the text attribute feature recognition method described in any one of Schemes 1 to 23;

[0308] The text relationship building module is used to establish an association relationship between the structured set and the phrase set according to the identified text attribute features.

[0309] Solution 44. A computer device comprising:

[0310] A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method for recognizing text attribute features as described in any one of Schemes 1-23, or the method for classifying text attribute features as described in any one of Schemes 24-26, or the method for constructing a classification model of text attribute features as described in Scheme 27, or the method for classifying text attribute features as described in Scheme 28 or 29, or the method for analyzing text structure as described in any one of Schemes 30-33 by executing the computer instructions.

[0311] Scheme 45. A computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, wherein the computer instructions are used to cause the computer to execute a method for identifying text attribute features as described in any one of Schemes 1-23, or to execute a method for classifying text attribute features as described in any one of Schemes 24-26, or to execute a method for constructing a classification model for text attribute features as described in Scheme 27, or to execute a method for classifying text attribute features as described in Scheme 28 or 29, or to execute a text structure analysis method as described in any one of Schemes 30-33.

[0312] The beneficial effect of the embodiments of the present invention is that by implementing the present invention, the text attribute features in the target text can be accurately identified, and the meaning of the text content can be recognized. Compared with the prior art, in addition to identifying the division of the text, parts of speech and other contents, it is also possible to accurately identify the characters, words, phrases, etc. that represent the text attribute features that represent the efficacy, effect and other contents in the text, and to explore the deeper meaning of the text. It can not only enrich the content of text recognition, but also provide more comprehensive data and content support for subsequent analysis and processing based on the text recognition content. BRIEF DESCRIPTION OF THE DRAWINGS

[0313] The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the present invention in any way. In the accompanying drawings:

[0314] Figure 1 A schematic diagram showing a flow chart of a method for identifying text attribute features according to an embodiment of the present invention;

[0315] Figure 2 A schematic diagram of the process of step S10 of an embodiment of the present invention is shown;

[0316] Figure 3 A schematic diagram of a directed graph G according to an embodiment of the present invention is shown;

[0317] Figure 4 A schematic diagram of a directed graph G according to another embodiment of the present invention is shown;

[0318] Figure 5 A schematic diagram showing the structure of a neural network model N according to an embodiment of the present invention is shown;

[0319] Figure 6 A schematic diagram showing the structure of the grammatical structure of an embodiment of the present invention;

[0320] Figure 7 A schematic diagram of the process of constructing the efficacy statement classification model according to an embodiment of the present invention is shown;

[0321] Figure 8 A schematic diagram of the process of constructing the efficacy phrase classification model according to an embodiment of the present invention is shown;

[0322] Fig. 9 A schematic diagram showing the structure of a syntax tree T according to an embodiment of the present invention is shown;

[0323] Fig.10 A schematic diagram showing a flow chart of a method for classifying text attribute features according to an embodiment of the present invention;

[0324] Fig.11 A schematic diagram showing the structure of a structure tree Tree according to an embodiment of the present invention is shown;

[0325] Fig.12 A schematic diagram of the process of constructing the engineering parameter classification model according to an embodiment of the present invention is shown;

[0326] Fig.13 A schematic diagram showing a flow chart of a text structure analysis method according to an embodiment of the present invention is shown;

[0327] Fig.14 A schematic diagram showing the structure of a device for recognizing text attribute features according to an embodiment of the present invention;

[0328] Fig.15 A schematic diagram showing the structure of a device for classifying text attribute features according to an embodiment of the present invention;

[0329] Fig.16 A schematic diagram showing the structure of a device for classifying text attribute features according to another embodiment of the present invention;

[0330] Fig.17 A schematic diagram showing the structure of a device for constructing a classification model of text attribute features according to an embodiment of the present invention;

[0331] Fig.18 A schematic diagram showing the structure of a text structure analysis device according to an embodiment of the present invention is shown;

[0332] Fig.19 A schematic diagram showing the hardware structure of a computer device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0333] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0334] As described in the background technology, the existing text recognition and analysis methods only stay at the level of identifying and analyzing the text itself, thereby analyzing the basic structure of the text, and cannot know the deeper meaning of the text. When users want to search based on a class of words that represent certain characteristics of the text (for example, words that express effects, words that express praise or derogatory meanings, etc.) as keywords, the existing text recognition and analysis methods cannot effectively identify such words, and therefore, the user's search needs are still not met. Language expressions are diverse, and keyword searches cannot quickly and effectively retrieve corresponding specific words.

[0335] Based on this, the embodiment of the present invention proposes a method for mining and displaying attribute features of text content. For any specified text content, such as a patent or a patent collection, the attribute feature content contained in the text content is mined, such as efficacy sentences and efficacy phrases that characterize efficacy, from which the actual value of each text content can be analyzed.

[0336] The embodiment of the present invention provides a method for identifying text attribute features, such as Figure 1 As shown, the identification method mainly includes the following steps:

[0337] Step S10: generating a grammatical structure according to the sentences in the target text;

[0338] Optionally, in some embodiments of the present invention, Figure 2 As shown, in step S10, the process of generating a grammatical structure according to a sentence in the target text mainly includes:

[0339] Step S11: Respectively identify the words of each sentence in the target text and construct a directed graph of words. In the embodiment of the present invention, a Chinese dictionary Dict is predefined, which consists of a common Chinese word library and a special word library. For the sentence sen in the target text, first identify each single word and each word in Dict, and then place these words in a directed graph G.

[0340] For example, the target text is "artificial intelligence technology". The directed graph G established based on the target text is as follows: Figure 3 As shown, each word in the sentence sen corresponds to an edge in G.

[0341] Step S12: Calculate the shortest path from the first node to the last node in the word directed graph as the word segmentation result of each sentence.

[0342] In constructing Figure 3 Following the directed graph G shown, with S1 as the starting node and S7 as the ending node, the routes corresponding to the possible word combinations in the sentence are constructed.

[0343] For example, if the dictionary contains the words "artificial", "intelligent", "technology" and "artificial intelligence", then the directed graph G becomes Figure 4 As shown. Figure 4 In the example, the edge "manual" points from S1 to S3, "intelligence" points from S3 to S5, "artificial intelligence" points from S1 to S5, and "technology" points from S5 to S7. If the weight of each edge is the same, the shortest path from S1 to S7 is S1-S5-S7, which contains only two edges, so the segmentation result is "artificial intelligence" + "technology".

[0344] In the embodiment of the present invention, the method for finding the shortest path may specifically be: arcs is defined as the adjacency matrix of the directed graph G. The matrix A is the path length between two vertices.

[0345] Initial conditions:

[0346] a) If there is an edge between node i and node j, then arcs[i][j] = the length of edge. Otherwise, arcs[i][j] = infinity.

[0347] b) In the 0th round of iteration, matrix A, A[i][j] = arcs[i][j].

[0348] Iterative algorithm:

[0349] A (k) [i][j]=min{A (k-1) [i][j],A (k-1) [i][k]+A (k-1) [k][j]}

[0350] Where k = 1, 2, 3, ..., n. The main idea is that if the length from node i to node k + the length from node k to node j is less than the length from node i directly to node j, then update the length from node i to node j.

[0351] Step S13: construct a word sequence according to the word segmentation results of each sentence; Step S14: generate an input vector according to adjacent words in the word sequence.

[0352] In the embodiment of the present invention, after word segmentation processing is performed on each sentence in the target text, a word sequence S after word segmentation is obtained, and the word segmentation result is input into the word2vec model training, the number of hidden layer nodes is 300, and a 300-dimensional vector of each word is obtained after training. A neural network N is defined, the first input layer is a 600-dimensional vector, the second hidden layer is a 300-dimensional vector, and the third output layer is a 600-dimensional vector.

[0353] For every two adjacent words term1 and term2 in the word sequence S, we get their corresponding vectors vec1 and vec2, which are concatenated into a 600-dimensional vector Vec, which is used as the input layer and output layer of the neural network model N. The reverse iterative algorithm is used to train the neural network model N. The structure of the neural network model N is as follows: Figure 5 shown.

[0354] Step S15: Obtain an output vector according to a preset neural network model and an input vector; Step S16: Calculate the cosine value of the angle between the input vector and the output vector.

[0355] In the embodiment of the present invention, for each sentence sen in the word sequence S, a syntax tree T is generated, and each word is a leaf node. The process of generating the syntax tree T mainly includes: for each two adjacent words term1 and term2 in sen, the corresponding vectors vec1 and vec2 are obtained, and they are concatenated into a 600-dimensional vector Vec_in, which is used as the input layer of the preset neural network model N to obtain the output layer vector Vec_out through calculation, and the angle cosine value cos1 between Vec_in and Vec_out is calculated.

[0356] Step S17: construct a combined node from two adjacent words with the largest angle cosine value until a root node of the word sequence is generated; Step S18: determine the grammatical structure of the word sequence based on the combined node and the root node.

[0357] For every two adjacent words in the target text, the cosine value of the angle is calculated through the above process, so as to determine the two adjacent words with the largest cosine value of the angle, add these two words into the middle node node of a syntax tree T, and obtain the vector v=vec1+vec2 of node. The above process is repeated until the root node root is found, thus forming the grammatical structure of the target text sen.

[0358] For example, the target text sen contains 5 words, term1 to term5, each corresponding to a leaf node. In the first round of iteration, the combined node node1 of term1 and term2 is calculated. In the second round of iteration, the combined node node2 of term3 and term4 is calculated. In the third round of iteration, the combined node node3 of node2 and term5 is calculated. In the last round, we only need to attribute node1 and node3 to the root, and finally form the following: Figure 6 The grammatical structure shown.

[0359] Step S20: Generate a data structure according to the node relationship in the grammatical structure;

[0360] For the target text, through the above step S10, its corresponding syntax tree T (i.e., grammatical structure) is determined, the root node is root, the intermediate nodes are n1, n2, ..., nk, and the leaf nodes are L1, L2, Lm. Two nodes with a parent-child relationship are generated into a data structure, for example: [node1→node2], thereby forming a feature list. For example, Figure 6 The grammatical structure shown in the figure has the corresponding feature list fea1 as follows: [root→node1], [root→node3], [node1→term1], [node1→term2], [node3→node2], [node3→term5], [node2→term3], [node2→term4].

[0361] Step S30: Generate a first input vector according to the data structure. In the embodiment of the present invention, the recognition of the text attribute features of the target text is mainly performed in combination with a pre-trained classification model. Therefore, after obtaining the data structure of the target text, an input vector of the classification model is formed according to the data structure.

[0362] In this data structure, the relationship between two words is also expressed. Therefore, the process of forming the input vector of the classification model based on this data structure mainly includes: inputting the words contained in the data structure into the word2vec model training, taking the number of hidden layer nodes as 300, and obtaining a 300-dimensional vector for each word after training. The first input layer of the classification model is defined as a 600-dimensional vector, the second hidden layer is a 300-dimensional vector, and the third output layer is a 600-dimensional vector.

[0363] For two related words in the data structure (such as [node2→term4] mentioned above), their corresponding vectors vec2 and vec4 are obtained and concatenated into a 600-dimensional vector Vec as the input vector of the classification model.

[0364] Step S40: determining the probability of each sentence containing attribute feature text according to the first input vector and a preset text attribute feature classification model.

[0365] After the input vector is obtained through the above step S30, the input vector is input into the text attribute feature classification model constructed through training, so as to determine the probability that each sentence in the target text contains the attribute feature text.

[0366] Step S50: Identify text attribute features in the target text according to probability.

[0367] After determining the probability that each sentence in the target text contains the attribute feature text, the probability can be used to identify whether the target text contains the text attribute feature. For example, if the probability that each sentence in the target text contains the attribute feature text is greater than a preset threshold, and the probability of each sentence in the target text containing the attribute feature text is determined, the sentences ranked in the top 10% are identified as containing the attribute feature text.

[0368] Through the text attribute feature recognition method of the embodiment of the present invention, the text attribute features in the target text can be accurately recognized, and the meaning of the text content can be recognized. Compared with the prior art, in addition to recognizing the division, part of speech and other contents of the text, it is also possible to accurately recognize the characters, words, phrases, etc. that represent the text attribute features that represent the efficacy, effect and other contents in the text, and to explore the deeper meaning of the text. It can not only enrich the content of text recognition, but also provide more comprehensive data and content support for subsequent analysis and processing based on the text recognition content.

[0369] Optionally, in some embodiments of the present invention, the target text may refer to a patent document or a patent collection, for example, and the text attribute feature corresponding to the patent document or patent collection may be, for example, an efficacy word characterizing the beneficial effect and technical effect of the patent. For example, for the sentence "achieving the effect of improving the running speed" in the target text, the efficacy word in the sentence is "improve", and other efficacy words may also be, for example, "easy to use, improve work efficiency, simple structure", etc. However, the efficacy word is only one type of text attribute feature, which is only an example and is not intended to limit the present invention.

[0370] Hereinafter, the solution of the embodiment of the present invention will be specifically described by taking patent documents or patent collections as the target text and efficacy words as the text attribute features as examples, but this is only an example and is not intended to limit the present invention.

[0371] Optionally, in some embodiments of the present invention, for the classification of efficacy statements, the above-mentioned preset text attribute feature classification model may be an efficacy statement classification model, such as Figure 7 As shown, the efficacy statement classification model can be constructed through the following process:

[0372] Step S71: Obtaining efficacy sentence samples, which include preset efficacy tags and sample sentences; efficacy sentence samples include sen (sentence) and efficacy tag flag (0 / 1 value, whether it is an efficacy sentence). For example, sample: Help users to reveal the obscure structure in the text, 1.

[0373] Step S72: generating a first grammatical structure according to the efficacy sentence sample;

[0374] Step S73: generating a first feature list according to the node relationship in the first grammatical structure;

[0375] Step S74: generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy mark and the sample sentence;

[0376] Step S75: training the preset classification model according to the first classification input vector and the first classification output vector to generate an efficacy statement classification model.

[0377] Specifically, a syntax tree T (i.e., grammatical structure) is extracted from the efficacy sentence sample, with the root node being root, the intermediate nodes n1, n2, ..., nk, and the leaf nodes L1, L2, and Lm. Two nodes with a parent-child relationship generate a data structure, [node1→node2], which constitutes a feature list as the input of the classification model. Whether the efficacy sentence is the output of the classification model, in an embodiment of the present invention, the classification model can select SVM, use a Gaussian kernel, and train to obtain the efficacy sentence classification model Model1.

[0378] Optionally, in some embodiments of the present invention, for the classification of efficacy phrases, the above-mentioned preset text attribute feature classification model may be an efficacy phrase classification model, such as Figure 8 As shown, the efficacy phrase classification model can be constructed through the following process:

[0379] Step S81: obtaining a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase;

[0380] Step S82: generating a second grammatical structure according to the functional phrase sample;

[0381] Step S83: generating a second feature list according to the node relationship in the second grammatical structure;

[0382] Step S84: generating a second classification input vector according to the feature list, and generating a second classification output vector according to the preset efficacy mark and the sample phrase;

[0383] Step S85: training the preset classification model according to the second classification input vector and the second classification output vector to generate a functional phrase classification model.

[0384] The first part of the efficacy phrase sample is a feature list of the sen grammatical structure. The second part is the composition of the phrase, qualifier + central word + efficacy word, such as "increase running speed", the central word is speed, the qualifier is running, and the efficacy word is improving. The second part of the efficacy phrase sample is in the form of speed | central word + running | qualifier + improving | efficacy word. The third part of the efficacy phrase sample is the efficacy tag, 0 / 1 value, indicating whether it is an efficacy phrase.

[0385] For example, a sample of efficacy phrase: achieve the effect of improving running speed, speed|central word+running|qualifier+improve|efficacy word, 1.

[0386] Extract the syntax tree T (syntactic structure) for the efficacy phrase sample, such as Fig. 9As shown, the root node is root, the intermediate nodes are n1, n2, nk. The leaf nodes are L1, L2, Lm. Two nodes with a parent-child relationship generate a data structure, [node1→node2], which constitutes a feature list. If term1, term2, term3 are efficacy phrases, then three words form a data structure, [term1, term2, term3], the first position is the central word, the second position is the qualifier, and the third position is the efficacy word. The two parts are combined as the input of the classification model, and the output is whether it is an efficacy phrase. The classification model can use SVM, use the Gaussian kernel, and train to obtain the efficacy phrase model Model2.

[0387] Optionally, in some embodiments of the present invention, the above step S40: the process of determining the probability that each sentence contains the attribute feature text according to the first input vector and the preset text attribute feature classification model may specifically include at least one of the following steps:

[0388] Determining a first probability that each sentence contains the attribute feature according to the first input vector and a preset text attribute feature classification model;

[0389] Determining a second probability that a phrase in each sentence contains an attribute feature according to the first input vector and a preset text attribute feature classification model;

[0390] A third probability that a paragraph in each sentence contains an attribute feature is determined according to the first input vector and a preset text attribute feature classification model.

[0391] In the embodiment of the present invention, the classification of the text attribute features of the target text may not be limited to the recognition of sentences, but may be extended to the recognition of phrases and paragraphs. Of course, those skilled in the art should be able to understand that the recognition method of the text attribute features in the embodiment of the present invention is not only for the sentences, phrases and paragraphs listed above, but may also include other text forms, and the present invention is not limited thereto.

[0392] Specifically, for patent documents, based on the above-mentioned functional sentence model Model1, each sentence in the patent document as the target text can be judged as a functional sentence to obtain the probability r1 that the corresponding sentence is a functional sentence; based on the above-mentioned functional phrase model Model2, functional phrases can be extracted from each sentence in the patent document as the target text to obtain the probability r2 of functional phrases.

[0393] For each phrase in the target text, the probability R that the phrase is an efficacy phrase is obtained, R=r1*w1+r2*w2+…, wherein w1, w2 are the weights of r1, r2, which can be set as needed, and the weighted part of the probability R of the efficacy phrase is not limited to the above-mentioned r1 and r2, but can also be extended to the probability r3 of the efficacy paragraph as needed.

[0394] Through the above calculation, if R is greater than a certain threshold and is in the top 10 of all phrases, then the phrase is an efficacy phrase. Those skilled in the art should know that the judgment criteria of R being greater than a certain threshold and being in the top 10 of all phrases described here are only examples, and in actual applications, the judgment criteria can be adjusted as needed.

[0395] Based on patent documents, identifying which specific words and phrases in the document belong to efficacy phrases helps to understand the actual role of the patent document. For example, patent documents basically have content describing the effects, whether in the invention content or in the specific implementation method. By identifying the description of the corresponding efficacy phrases and quickly locating the effect part, you can have a more accurate understanding of the beneficial effects of the scheme involved in the patent document. Moreover, in actual applications, based on the identified efficacy phrases, you can also add indexing tags to the patent document, which can give searching users an additional dimension to understand the patent document.

[0396] For a patent collection, each sentence of each patent in the patent collection can be judged as an efficacy sentence, and the probability r1 of this sentence being an efficacy sentence can be obtained. The efficacy phrase can be extracted from each sentence of the patent, and the probability r2 of the efficacy phrase can be obtained.

[0397] For each phrase in the target text, the probability R that the phrase is an efficacy phrase is obtained, R=r1*w1+r2*w2, w1, w2 are the weights of r1, r2, which can be set as needed, and the weighted part of the probability R of the efficacy phrase is not limited to the above-mentioned r1 and r2, but can also be extended to the probability r3 of the efficacy paragraph as needed.

[0398] Through the above calculation, if R is greater than a certain threshold and belongs to the top 10% of all phrases, then the phrase is an efficacy phrase. Those skilled in the art should know that the judgment criteria of R being greater than a certain threshold and belonging to the top 10% of all phrases described here are only examples. In actual applications, the judgment criteria can be adjusted as needed.

[0399] After the corresponding efficacy phrases are identified, the efficacy phrases appearing in the patent collection are also integrated to obtain the efficacy phrase set of the patent collection. Based on the efficacy phrase set, a macroscopic understanding of the overall situation of the patent collection can be obtained. In practical applications, the corresponding patent collection is usually constructed based on the core of fields, technical directions, keywords, etc., or the corresponding patent collection is presented based on the content retrieved by the user. For this type of patent collection, the efficacy phrases existing in a class of patent collections can be effectively identified through the recognition method of the text attribute features of the embodiment of the present invention, and subsequent statistical analysis and other processing can be performed based on the recognition results.

[0400] It should be noted that, as mentioned above, in the embodiments of the present invention, the target text is not limited to patent documents or patent collections, but the corresponding content can be identified for various texts, and the identified content is not limited to sentences representing efficacy, but can be other types of characters, words, sentences, etc. that represent text attribute characteristics, and the present invention is not limited to this.

[0401] The text attribute feature recognition method of the embodiment of the present invention is further described below in conjunction with some specific application examples.

[0402] In actual applications, when users want to search for text content such as patent documents, they often use search formulas composed of relevant core words such as keywords, technical fields, and classification numbers to search. However, due to the limitations of the search formula, the search results obtained are often large in number and the search scope is large. At this time, if further screening can be performed in combination with the specific efficacy and beneficial effects of the patent documents, the scope of the retrieved content can be effectively further narrowed, thereby providing users with more accurate search results.

[0403] Therefore, the text attribute feature recognition method provided by the embodiment of the present invention can further identify the functional content of text content such as patent documents, so as to be used by users to screen the text content. For example, the content retrieved by the user contains the functional description of "electric stove + increase + water + temperature". At this time, through the text attribute feature recognition method described in any of the above embodiments, the above functional description in the patent document can be identified, wherein water is the central word (i.e., the object of action); increase is the functional word (i.e., the action applied to the object of action), and it is also a word that indicates the positive or negative effect of the efficacy, which can be specifically defined as a directional word; temperature is a parameter; electric furnace is a qualifier (in this example, the electric furnace is a functional carrier). Among them, qualifiers refer to words that limit the core concept of efficacy; central words refer to the core concept of technical efficacy; directional words refer to words that describe the optimization direction of technical effects.

[0404] Therefore, based on the basic search, the user can select the literature content related to "electric stove + increase + water + temperature" according to the screening of further functional descriptions. Of course, in actual applications, the functional descriptions in the literature may not be so specific. For example, it can be more generalized such as "increase + water + temperature", "heat water", "increase temperature", "electric stove heats water", or only contain part of the functional description. However, no matter how the form of the functional description changes, the form of the central word + efficacy word it contains will not change. Therefore, for this type of functional description, the text attribute feature recognition method of the embodiment of the present invention can recognize it, so as to provide users with selection, screening and other processing.

[0405] The embodiment of the present invention also provides a method for classifying text attribute features, such as Fig.10 As shown, the classification method mainly includes:

[0406] Step S101: Identify the text attribute features in the target text according to the sentences in the target text. In the embodiment of the present invention, the classification premise of the classification method of text attribute features is to identify the text attribute features existing in the sentences of the target text, so as to classify them on this basis. Optionally, in some embodiments of the present invention, the marked characters, words, phrases, etc. can be identified by indexing or highlighting to obtain the corresponding text attribute features. In some embodiments of the present invention, the target text can also be identified by the text attribute feature identification method described in any of the above embodiments to obtain the recognition result of the text attribute features.

[0407] Step S102: Construct a text attribute feature set based on the recognition result of the target text. The recognized text attribute features usually also include the corresponding core words. For example, as mentioned above, the word combination obtained includes not only the efficacy words, but also the core words and qualifiers. For such a word combination, a text attribute feature set T is defined, which includes n elements. For example: t1 = operation + speed, there are n such combinations t1 in the whole text, thus forming a set T.

[0408] Step S103: construct a structure tree based on the text attribute feature set. Based on the text attribute feature set T, a structure tree Tree is established, such as Fig.11 As shown, the root is the root node, and each element in T is a leaf node of the Tree. The leaf nodes are combined with L, including L1, L2, and Ln. If any two leaf nodes Li and Lj have co-occurred in S, Li and Lj are included in an intermediate node. The upper layer node of the leaf node is Node, including Node1, Node2, and Nodek. All Node nodes are connected to the root.

[0409] Step S104: respectively determine the class center point of each leaf node in the structure tree, and form a class center set with multiple class center points.

[0410] In the embodiment of the present invention, this process is to use one of the leaf nodes in the structure tree as the central node, calculate the average distance from the leaf node to the central node in the structure tree; and determine the leaf node with the farthest average distance as the class center.

[0411] Step S105: Generate a class center vector according to multiple class center points, and generate a leaf node vector according to the leaf nodes of the structure tree; Step S106: Determine the classification node of each leaf node according to the class center vector and the leaf node vector classification node.

[0412] In an embodiment of the present invention, the process of determining the classification node of each leaf node based on the class center vector and the leaf node vector classification node mainly includes: calculating the Euclidean distance from the leaf node vector to the class center vector; for each leaf node, taking the class center point corresponding to the class center vector with the smallest Euclidean distance as the classification node of each leaf node.

[0413] The above process is to classify the above n elements (i.e. the identified text attribute features) into m clusters (for example, m=10000 can be defined). Specifically, the classification process is as follows:

[0414] (1) Initialize the class center set C, the initial state C is empty;

[0415] (2) Randomly select a leaf Li and add it to set C;

[0416] (3) For other nodes, calculate the leaf Lj with the longest average distance from the leaf node Li in set C, where the distance is equal to the number of edges that a node passes through when walking through the tree to reach Lj. Add the leaf Lj with the longest average distance to set C.

[0417] (4) Repeat step (3) until there are m elements in set C, that is, m cluster centers, and cluster_i represents a center point.

[0418] For each cluster_i, it is composed of a qualifier and a central word, and the qualifier vector and the central word vector are obtained and concatenated into a 600-dimensional vector vector_i.

[0419] For each leaf node leaf, we also get a 600-dimensional vector vec concatenated with the leaf's qualifier word vector and the center word vector. We calculate the Euclidean distance r from vec to each cluster_i vector, and take the center point with the smallest r as the leaf's classification.

[0420] Through the above process, the corresponding class center can be found for each leaf. The class center phrase is the classification result of each text attribute feature in the class.

[0421] For example, after the calculation in the above process, the running speed of L1 is a class center. The sliding speed of L2 is closest to L1, so L1 is the classification result of L2.

[0422] Furthermore, in some embodiments of the present invention, the cosine value of the angle between central phrases of different classes can be calculated (for example, calculated through the process of steps S11 to S17 above), and if the cosine value of the angle between two central phrases of different classes is greater than a preset threshold (for example, 0.95), it can be determined that the meanings expressed by the two central phrases of different classes are synonymous or near-synonymous, and the two central phrases of different classes can be further classified to obtain further classification results.

[0423] In actual applications, the functional descriptions contained in the content retrieved by the user may exist in various forms, but their meanings may belong to the same type of functions or the effects that can be achieved may belong to the same type. For example, "electric stove + increase + water + temperature", "increase + water + temperature", "heat water", "raise temperature", "electric stove heats water", the substantive contents of these functional descriptions are actually the same type, and are all used to increase the temperature of water, and the difference lies only in the slightly different forms of expression. Therefore, the functional descriptions can be classified by the classification method of the text attribute features of the embodiment of the present invention, so as to realize the classification of text contents that are only different in the form of expression, such as the above-mentioned, and to divide the words describing functions and effects with different substantive contents, and summarize the corresponding set of effect phrases. At the same time, it can also avoid dividing the functional descriptions with the same substantive content into separate categories, which increases the complexity of the classification results.

[0424] Based on the above classification process, at least two problems can be solved. On the one hand, in practical applications, users can also search based on the efficacy description when searching. However, from the user's perspective, users hope to quickly search all the expressions related to the efficacy in a more comprehensive way without listing all the expressions with the same substantive content. From the above examples, it can be seen that for the functional descriptions with the same actual meaning, there may be many different expressions, which the user may not be able to list. Even if the user lists them, the entire search process will become very cumbersome. Therefore, how to simplify this search process should be focused on in the search scheme based on text content. Therefore, based on the above classification method, the functional descriptions with the same substantive meaning are classified. Based on the classification results, when the user searches, only one of the expressions needs to be entered. The database, document library, etc. used by the user can identify the classification of the functional description that the user wants to search based on the functional description classification to which the expression belongs, and extract the corresponding search results from the database and document library based on the classification. It can be seen that such a processing method can greatly simplify the user's operation process, and at the same time can obtain more comprehensive search results that meet the user's needs.

[0425] On the other hand, the classification results obtained through the above process can be applied to scenarios such as databases and document libraries. When users retrieve corresponding results based on search terms such as keywords, they can classify them based on the functional descriptions in the preliminary search results, thereby refining the main types of functional descriptions contained in the preliminary search results. Based on these contents, a filtering condition can be added to the filtering conditions of the database and document library, and the classified functional descriptions can be used as filtering items for users to filter, providing users with more dimensional search filtering conditions, thereby helping users to locate the scope or content they want to search more quickly and accurately.

[0426] At this time, the above-mentioned functional description existing in the patent document can be identified by the recognition method of the text attribute feature described in any of the above embodiments, wherein water is the central word (i.e., the object of action), increase is the efficacy word (i.e., the action applied to the object of action), temperature is a parameter, and the electric furnace is a qualifier (in this example, the electric furnace is the functional carrier). Therefore, on the basis of basic search, the user can select the document content related to "electric furnace + increase + water + temperature" according to the screening of further functional descriptions. Of course, in practical applications, the functional description existing in the document may not be so specific. For example, it can be a more generalized functional description of this type, or only contain a part of it. However, no matter how the form of the functional description changes, the form of the central word + efficacy word contained therein will not change. Therefore, for this type of functional description, the recognition method of the text attribute feature of the embodiment of the present invention can be used for identification, thereby providing the user with selection, screening and other processing.

[0427] Optionally, in some embodiments of the present invention, when classifying text attribute features, a situation may be encountered, that is, one type of words expresses a "positive" meaning, while another type of words expresses a "negative" meaning, but the actual meanings of the two are actually the same. For example, the actual meanings of "increase + accuracy" and "reduce + error rate" are the same. However, "increase + accuracy" belongs to "positive" words, and "reduce + error rate" belongs to "negative" words. If only one type of words is classified, some words will obviously be lost. Classifying these two different types of words separately is somewhat inconsistent with the fact that their actual meanings are the same. Therefore, in an embodiment of the present invention, in order to avoid this situation, when classifying, the words with the same actual meanings can also be normalized.

[0428] For m types of central phrases, a cognitive normalization rule library can be established by indexing. For central words and text attribute features (in this embodiment, efficacy words are used as an example for illustration). Antonym libraries are established separately: for central words, for example, error rate-accuracy rate. For each central word in the m phrases, the central word with the opposite meaning is obtained, and each negative central word is replaced with a positive central word. For efficacy words, for example, reduce-increase. For each efficacy word in the m phrases, the efficacy word with the opposite meaning is obtained, and each negative efficacy word is replaced with a positive efficacy word.

[0429] Through such processing, two types of words with the same actual meaning are unified. When identifying and classifying the target text, the corresponding results can be more accurate and more in line with the actual meaning expressed by the text content.

[0430] In practical applications, the above normalization process can be carried out by learning and training the same or different semantic relationships such as positive and negative effect words through, for example, intelligent semantic analysis technology in artificial intelligence, so as to form a semantic analysis model that can classify the input words. The semantic analysis model can be used to realize fast and efficient normalization of a large number of words. The normalization process combined with intelligent semantic analysis mainly includes:

[0431] 1. Through the semantic analysis model, the effect categories in the target text are identified and divided into positive effect phrases (such as words such as improve and enhance) and negative effect phrases (such as words such as reduce and decrease).

[0432] Among them, the process of training the semantic analysis model based on efficacy category words mainly includes:

[0433] A) Establish a corpus. For example, 10,000 positive corpora and 10,000 negative corpora can be selected. Each corpus follows the format of qualifier + central word + efficacy word. For example, "increase running speed", the central word is speed, the qualifier is running, and the efficacy word is improvement. The corresponding corpus structure is: speed | central word + running | qualifier + improvement | efficacy word. The category represents a 0 / 1 value, 0 is negative, and 1 is positive. Therefore, for the positive efficacy phrase "increase running speed" and the negative efficacy phrase "reduce running speed", the corresponding corpus can be obtained as follows:

[0434] speed|central word+operation|qualifier+improvement|efficacy word, 1;

[0435] speed|central word+operation|qualifier+reduce|efficacy word,0.

[0436] B) Use the SVM algorithm to classify each efficacy phrase into a category, use the Gaussian kernel, and train based on the corpus to obtain a semantic analysis model.

[0437] 2. Convert the identified negative phrases into positive phrases

[0438] A) Establish an antonym word library, such as error rate-correct rate. For each central word in the m phrases, calculate the central word with the opposite meaning, and replace each negative central word with a positive central word.

[0439] B) Create an antonym word library, such as lower-raise. For each function word in the m phrases, obtain the function word with the opposite meaning, and replace each negative function word with a positive function word.

[0440] At the same time, in the above examples, only the process of normalizing positive and negative efficacy words is used to illustrate. In an optional embodiment of the present invention, the normalization of efficacy words is not limited to semantically opposite efficacy words, and can also be normalized for words with similar semantics (such as synonyms or near-synonyms). For example, for words such as "improve, enhance, increase, increase", they are all synonyms or near-synonyms that represent the same meaning. Based on their same meaning, one of the efficacy words can be used as the basis to normalize other synonyms or near-synonyms. Through this normalization process, the accuracy of classification based on efficacy words can be further improved. At the same time, in the subsequent retrieval, classification or screening process, based on the screening words after normalization, the results obtained by the user after the selection based on the screening words can also be more in line with the user's needs, thereby improving the user's experience.

[0441] In the process of combining various methods of the embodiments of the present invention for practical application, it is found that the text attribute characteristics of the target text targeted by the recognition method and classification method of the embodiments of the present invention can also have the same content as the currently popular TRIZ theory (literally translated as the theory of solving invention problems. TRIZ theory successfully reveals the inherent laws and principles of creative inventions, focuses on clarifying and emphasizing the contradictions existing in the system, and its goal is to completely resolve the contradictions and obtain the ultimate ideal solution). Therefore, in the embodiments of the present invention, the definition of the engineering parameter set in TRIZ theory is used to identify or classify the text attribute characteristics of the target text.

[0442] Among them, TRIZ's definition of engineering parameter set can be shown in Table 1:

[0443] Table 1

[0444]

[0445] As can be seen from Table 1, in TRIZ theory, there is already a certain classification of efficacy parameters, which is consistent with the concept of efficacy parameters used as examples in the embodiments of the present invention. In addition, in TRIZ theory, there is also a certain division of physical parameters, which is also consistent with the text attribute characteristics of the target text targeted in the embodiments of the present invention.

[0446] Therefore, an embodiment of the present invention provides a classification method for text attribute features, which mainly includes:

[0447] First, the target text is obtained and the text attribute features of the target text are identified. In the embodiment of the present invention, the classification premise of the classification method of text attribute features is to identify the text attribute features existing in the sentences of the target text, so as to classify them on this basis. Optionally, in some embodiments of the present invention, the marked characters, words, phrases, etc. may be identified by indexing or highlighting to obtain the corresponding text attribute features. In some embodiments of the present invention, the target text may also be identified by the text attribute feature identification method described in any of the above embodiments to obtain the recognition result of the text attribute features.

[0448] Then, the engineering parameter classification result corresponding to the text attribute feature is generated according to the text attribute feature and the preset engineering parameter classification model. The preset engineering parameter classification model is a neural network model constructed by pre-training according to the classification of engineering parameters in the TRIZ theory described above. Therefore, based on the trained classification model, the result corresponding to the engineering parameter classification can be obtained based on the text attribute feature.

[0449] Optionally, in some embodiments of the present invention, Fig.12 As shown, the engineering parameter classification model can be trained and constructed through the following process:

[0450] Step S121: Acquire engineering parameter corpus samples and text attribute feature samples, wherein the engineering parameter corpus samples include engineering parameter text attribute features and text attribute feature type tags, for example, for "increase weight, 1", where increase weight is an engineering parameter text attribute feature, 1 corresponding type tag indicates that increase weight belongs to the engineering parameter text attribute feature, and if the tag is 0, it indicates that the corresponding phrase does not belong to the engineering parameter text attribute feature. In an optional embodiment, the text attribute feature samples described herein may be generated by identifying the text attribute feature by the method for identifying the text attribute feature described in any of the above embodiments;

[0451] Step S122: Generate an engineering parameter input vector based on the text attribute feature sample, and generate an engineering parameter output vector based on the engineering parameter corpus sample. Based on the above samples, the input vector and output vector required for the training model are constructed.

[0452] Step S123: Train the SVM model according to the engineering parameter input vector and the engineering parameter output vector to construct the engineering parameter classification model. In the embodiment of the present invention, each text attribute feature phrase can be classified into an engineering parameter text attribute feature using an SVM algorithm, and a Gaussian kernel can be used to train a neural network model to construct the desired engineering parameter classification model.

[0453] Through the above method, combined with the relatively mature TRIZ theory, the words that represent the attribute characteristics of the text are classified accordingly, and based on this classification method, the target text can be further analyzed and mined, and classified based on the engineering parameters of the TRIZ theory, so that the target text has a more systematic classification method. In addition, with the widespread application of TRIZ theory in various fields, the content displayed by the text classification results combined with TRIZ theory can better meet the needs of users and provide users with important reference content for mining, analysis, and expansion plans in the corresponding fields.

[0454] Optionally, in some embodiments of the present invention, in combination with the classification method of the text attribute features described in the above embodiments, when classifying the text content, in addition to the classification based on the engineering parameters of TRIZ theory, the functional classification of the engineering parameters in TRIZ theory can also be combined to classify. In TRIZ theory, the words corresponding to the engineering parameters can be divided into useful functions, harmful functions, insufficient functions and transition functions. The classification results in this regard can be considered from two levels. On the one hand, it can be divided only according to which category of these functional classifications the words belong to, without relying on the engineering parameter classification described above; on the other hand, it can also be further divided according to which category of function the words belong to on the basis of the engineering classification described above. The specific classification method can be set according to actual needs. For example, for users to choose different classification methods, you can choose to directly classify according to the function; and for users to choose to classify the words according to different classification levels, you can choose to further classify the functions based on the engineering parameter classification, but the present invention is not limited to this.

[0455] Moreover, in practical applications, the text attribute feature analysis method of engineering parameters combined with TRIZ theory can also be used in combination with the text attribute feature analysis method described in some embodiments of the present invention. For the content that the user wants to retrieve, not only one aspect of the analysis and retrieval results are listed, but the retrieval results classified based on general functional words can be displayed, and the retrieval results classified based on engineering parameters of TRIZ theory can also be displayed at the same time, so as to display the corresponding retrieval results to the user through different dimensions and different levels for the text content that the user wants to retrieve, and display more comprehensive and detailed retrieval results for the user. For such a retrieval scheme, in practical applications, for example, an "automatic functional word composite retrieval function" can be added to the database, document library, etc. For example, when the user selects the automatic functional word to meet the retrieval function, and enters the keyword "anti-loosening" that he wants to retrieve in the corresponding input field, the system of the database and document library can retrieve the words of the same functional description with the same substantive meaning based on the keyword, and based on the corresponding classification, retrieve the keywords related to it in the technical effect and TRIZ engineering parameters, as well as the technical effect sentences and technical effect phrases corresponding to these keywords. The user selects the function sentence, function phrase or function word related to the function to be searched from the "corresponding technical function sentences and technical function phrases" found. Correspondingly, if the keyword that the user inputs to search is the above-mentioned "related function sentence, function phrase or function word", the database and document library system can also reversely search for the corresponding technical function level 1 to 3 and triz parameters based on the content.

[0456] For the search results obtained in the above process, users can further filter the above information according to their own needs and the filter items listed by the system to find the text content they really want. In actual application, users can use OR statements to combine the above search keywords into a review search formula to conduct a comprehensive search for the content related to the efficacy.

[0457] The embodiment of the present invention also provides a text structure analysis method, such as Fig.13 As shown, the text structure analysis method mainly includes:

[0458] Step S131: Identify the phrases in each paragraph of the target text, and form a phrase set according to each paragraph. For any target text (such as patent document d), identify each paragraph in the target text, and form a phrase set S for the phrases that appear in the same paragraph. For example, patent document d has k paragraphs, p_1 to p_k, and the phrase set S contains k phrase sets S1, S2, ..., Sk.

[0459] Step S132: Perform structural analysis on each phrase set to generate a structured set. Perform structural analysis on each paragraph of patent document d to obtain a set STRUCT, which contains k elements, each of which is a structured result of a paragraph.

[0460] Step S133: Identify the text attribute features in the paragraphs of the structured set. In the embodiment of the present invention, the classification premise of the classification method of the text attribute features is to identify the text attribute features existing in the sentences of the target text, and then classify them based on this. Optionally, in some embodiments of the present invention, the marked characters, words, phrases, etc. can be identified by indexing or highlighting to obtain the corresponding text attribute features. In some embodiments of the present invention, the target text can also be identified by the text attribute feature identification method described in any of the above embodiments to obtain the recognition result of the text attribute features.

[0461] Step S134: establishing an association relationship between the structured set and the phrase set according to the identified text attribute features.

[0462] In the embodiment of the present invention, if the paragraph p_i corresponding to struct_i contains the text attribute characteristic phrase set s_i, a corresponding relationship between struct_i and S_i is established.

[0463] If the paragraph p_i corresponding to struct_i does not contain the text attribute feature phrase, then the phrase set S_j that is closest to p_i and appears after p_i is selected.

[0464] If S_j does not exist, select the phrase set S_m that is closest to p_i and appears before p to establish a corresponding relationship.

[0465] If S_m still does not exist, p_i is ignored.

[0466] Through the above process, the corresponding relationship between the extraction structure and efficacy of patent document d is determined.

[0467] For example, for the target text:

[0468] (p1) 1. The gantry moves longitudinally through the lead screw and the connecting block, driving the tool to move longitudinally. This coordination method runs stably and improves engraving accuracy;

[0469] (p2) Second, there are multiple centers on the center rack, and through the cooperation of guide sleeves and guide pillars, you only need to push the center rack to clamp multiple workpieces at the same time, and tighten the tightening bolts on the guide sleeve to fix the center rack. It is easy to operate and improves efficiency.

[0470] It can be seen that the target text has two paragraphs, p1 and p2.

[0471] For paragraph p1:

[0472] 1. Through the above steps S131 and S132, a structure struct_1 is obtained: gantry-connection-screw, gantry-drive-tool;

[0473] 2. Based on the above steps S131 and S132, combined with step S133, obtain the functional phrase s_1: stable operation and improved engraving accuracy;

[0474] Based on the above steps S131 to S133 , combined with step S134 , it is determined that the association relationship between struct_1 and s_1 is such that the effect of s_1 is achieved through struct_1 .

[0475] For paragraph p2:

[0476] 1. Through the above steps S131 and S132, a structure struct_2 is obtained: top rack-including-top rack, top rack-connecting-guide sleeve, top rack-connecting-guide column, top rack-pushing-workpiece;

[0477] 2. Based on the above steps S131 and S132, combined with step S133, obtain the functional phrase s_2: convenient operation and improved efficiency;

[0478] Based on the above steps S131 to S133, combined with step S134, the corresponding association relationship between struct_2 and s_2 is to achieve the effect of s_2 through struct_2.

[0479] Furthermore, for a target text set containing multiple target texts (for example, a patent set D containing multiple patent documents), firstly, the phrase set of each patent is obtained through the above process, and the phrase set is structured, and the correspondence between struct_i and the text attribute feature phrase is determined. For each phrase set s, all the corresponding structures in the patent set D are obtained. For each structure struct, all the corresponding efficacy phrases in the patent set D are obtained. In this way, the correspondence between the extraction structure and efficacy in the patent set D is completed.

[0480] Through the above process, it is possible to recommend a structured technical solution for the analysis of text attribute features, or to predict the text attribute feature content of the structured technical solution). Moreover, it is also possible to combine certain restrictions to give more accurate results. For example, by combining efficacy and technical fields or keywords, a targeted technical solution can be given, or a more accurate efficacy prediction can be given for a certain structured technical solution and key components (multiple prediction results can be given, and the probability can be expressed as a percentage).

[0481] The above-mentioned various method embodiments are respectively mining, analyzing, identifying, clustering and other processes of text attribute features based on text from the perspectives of text attribute feature recognition method, text attribute feature classification method, text structure analysis method, etc. In practical applications, they can achieve their respective functions, obtain corresponding analysis and processing results, and provide the results to users. Optionally, in some embodiments of the present invention, the methods corresponding to the above-mentioned various method embodiments can also be combined to obtain an overall mining, analysis and other processing processes for text, so as to show users different results that can be obtained based on the text from multiple angles and multiple levels.

[0482] In an embodiment of the present invention, the schemes corresponding to different method embodiments may constitute an overall analysis and processing architecture for text, which may include processing results at multiple different levels, such as: language layer, perception layer, cognitive layer, and standard parameter layer.

[0483] The language layer is used to represent the part of speech marking and splitting of efficacy words, efficacy phrases, etc., which reflects the process of identifying the text attribute features of efficacy words, efficacy phrases, etc. The language layer can be a combination of qualifiers, central words, and direction words. The language layer can be obtained by the text attribute feature identification method described in the above method embodiment.

[0484] The perception layer is used to represent the result of classifying the efficacy words in the text, which is to standardize and unify the words with the same meaning, and to combine the standardized words. The perception layer can be obtained by the classification method of the text attribute features of the above method embodiment; the ability to generalize from the language layer to the perception layer is equivalent to the ability of computers to achieve human perception of technical efficacy.

[0485] The cognitive layer is used to characterize the result of normalizing the expression of the efficacy words in the text. On the basis of the perception layer technical efficacy phrases, the qualifiers, central words and directional words are taken as a whole, and the consistency of people's cognition of the substantive content of the language expression is used as the standard for unification and standardization. The cognitive layer can be obtained through the normalization process in the classification method of the text attribute features of the above method embodiment. The ability to generalize from the perception layer to the cognitive layer is equivalent to the ability of computers to achieve people's cognitive ability of the substantive content of technical efficacy.

[0486] The standard parameter layer is used to characterize the result of clustering the technical efficacy of the cognitive layer, matching the functional parameters of the technical contradiction matrix of TRIZ theory, and forming the efficacy TRIZ parameters. The standard parameter layer can be obtained by the classification method of the above method embodiment combined with the text attribute characteristics of TRIZ theory. The ability to summarize from the perception layer to the TRIZ parameter layer is equivalent to the ability of computers to summarize the technical efficacy of invention experts.

[0487] In practical applications, based on the processing of the above-mentioned method embodiments, corresponding analysis and processing results can be generated for the text to be retrieved by the user for the user to choose, or different analysis results can be displayed according to the user's screening, and the corresponding retrieval analysis results of the text retrieved by the user can be displayed from different angles and in different ways. The following is a description of the processing results of the analysis and processing architecture constructed based on the method embodiments of the present invention in combination with several specific embodiments.

[0488] Example 1

[0489] Text 1: The beneficial effects of the device are that it is easy to install and the detection error rate is reduced.

[0490] Two efficacy phrases are extracted from text 1.

[0491] Function phrase 1 of text 1: Easy to install

[0492]

[0493] Effect phrase 2 of text 1: detection error rate is reduced

[0494]

[0495]

[0496] Text 2: The device is easy to install and has high detection accuracy.

[0497] Function phrase 1 of text 2: Easy installation

[0498]

[0499] Functional phrase 2 of text 2: High detection accuracy

[0500]

[0501] Example 2 is an example of the result of normalizing the expression of the efficacy words in the text implemented by the above-mentioned cognitive layer:

[0502] Text 1: The beneficial effects of the device are that it is easy to install and the detection error rate is reduced.

[0503] Text 2: The device is easy to install and has high detection accuracy.

[0504] It can be seen that "detection error rate reduction" in text 1 and "detection accuracy is high" in text 2 are two types of efficacy words with the same actual meaning. When a user searches for text that can improve detection accuracy, the user's input language may be "how to improve detection accuracy", "technology with low detection error rate", etc. If technical efficacy is not normalized, text 1 and text 2 cannot be matched successfully.

[0505] After the efficacy is extracted by the method of the embodiment of the present invention, the cognitive layer results are as follows:

[0506]

[0507] It can be seen that after normalizing the user input requirements, the two questions are "improved detection accuracy" at the cognitive level, and the substantive content of text 1 and text 2 at the cognitive level is completely consistent. Both text 1 and text 2 are successfully matched, and the comprehensiveness of the retrieval is improved compared to the unnormalized technical efficacy.

[0508] Example 3 is an example of the result of normalizing the expression of the efficacy words in the text implemented by the above-mentioned cognitive layer:

[0509] Text 4: The water dispenser is easy to install.

[0510] Text 5: Easy-to-use water dispenser.

[0511] User input question: Easy to install water dispenser.

[0512] From the perspective of sentence structure and text similarity, compared with text 4, text 5 is closer to the user input content. If it is based on a traditional search engine, the similarity score of text 5 may be higher than that of text 4. From a human perspective, the problem of text 4 and user input is to clarify the description of the effect to the convenience of installation, which is different from the convenience of the "use" process of text 5.

[0513] After the efficacy is extracted by the method of the embodiment of the present invention, the cognitive layer efficacy and matching results are as follows:

[0514]

[0515]

[0516] Therefore, based on the results obtained by the method of the embodiment of the present invention, compared with text 5, the cognitive layer technical efficacy of text 4 is more closely matched with the cognitive layer technical efficacy input by the user.

[0517] The embodiment of the present invention also provides a device for identifying text attribute features, such as Fig.14 As shown, the text attribute feature recognition device mainly includes:

[0518] The grammatical structure generating module 141 is used to generate a grammatical structure according to the sentences in the target text; the details can be found in the relevant description of step S10 of the above method embodiment, which will not be repeated here;

[0519] The data structure generating module 142 is used to generate a data structure according to the node relationship in the grammatical structure; the details can be found in the relevant description of step S20 of the above method embodiment, which will not be repeated here;

[0520] A first input vector generating module 143 is used to generate a first input vector according to the data structure; the details can be found in the relevant description of step S30 in the above method embodiment, which will not be repeated here;

[0521] The probability determination module 144 is used to determine the probability that each sentence contains the attribute feature text according to the first input vector and the preset text attribute feature classification model; the details can be found in the relevant description of step S40 of the above method embodiment, which will not be repeated here;

[0522] The text recognition module 145 is used to recognize text attribute features in the target text according to probability; the details can be found in the relevant description of step S50 of the above method embodiment, which will not be repeated here.

[0523] Through the text attribute feature recognition device of the embodiment of the present invention, the text attribute features in the target text can be accurately recognized, and the meaning of the text content can be recognized. Compared with the prior art, in addition to recognizing the division, part of speech and other contents of the text, it is also possible to accurately recognize the characters, words, phrases, etc. that represent the text attribute features that represent the efficacy, effect and other contents in the text, and can explore the deeper meaning of the text. It can not only enrich the content of text recognition, but also provide more comprehensive data and content support for subsequent analysis and processing based on the text recognition content.

[0524] The embodiment of the present invention also provides a text attribute feature classification device, such as Fig.15 As shown, the classification device of the text attribute feature mainly includes:

[0525] The text recognition module 151 is used to recognize text attribute features in the target text according to the sentences in the target text; the details can be found in the relevant description of step S101 in the above method embodiment, which will not be repeated here;

[0526] The text attribute feature set construction module 152 is used to construct a text attribute feature set according to the recognition result of the target text; the details can be found in the relevant description of step S102 of the above method embodiment, which will not be repeated here;

[0527] A structure tree construction module 153 is used to construct a structure tree according to the text attribute feature set; the details can be found in the relevant description of step S103 of the above method embodiment, which will not be repeated here;

[0528] The class center set construction module 154 is used to determine the class center point of each leaf node in the structure tree respectively, and form a class center set with multiple class center points; the details can be found in the relevant description of step S104 of the above method embodiment, which will not be repeated here;

[0529] A vector generating module 155 is used to generate a class center vector according to the plurality of class center points, and to generate a leaf node vector according to the leaf nodes of the structure tree; for details, please refer to the relevant description of step S105 in the above method embodiment, which will not be repeated here;

[0530] The classification node determination module 156 is used to determine the classification node of each leaf node according to the class center vector and the leaf node vector classification node; for details, please refer to the relevant description of step S106 of the above method embodiment, which will not be repeated here.

[0531] By using the text attribute feature classification device of the embodiment of the present invention to classify the function description, it is possible to classify the text contents that differ only in the form of expression, for example, and to classify the words describing functions and effects with different substantial contents, and summarize the corresponding effect phrase sets. At the same time, it is also possible to avoid classifying the function descriptions with the same substantial contents into separate categories, which increases the complexity of the classification results.

[0532] The embodiment of the present invention also provides a text attribute feature classification device, such as Fig.16 As shown, the classification device of the text attribute feature mainly includes:

[0533] The text recognition module 161 is used to obtain the target text and identify the functional phrases of the target text. In the embodiment of the present invention, the classification premise of the classification method of the text attribute feature is to identify the text attribute features existing in the sentences of the target text, so as to classify them based on this. Optionally, in some embodiments of the present invention, the marked characters, words, phrases, etc. can be identified by indexing or highlighting to obtain the corresponding text attribute features. In some embodiments of the present invention, the target text can also be identified by the text attribute feature identification method described in any of the above embodiments to obtain the recognition result of the text attribute feature.

[0534] The text classification module 162 is used to generate the engineering parameter classification result corresponding to the efficacy phrase according to the text attribute features and the preset engineering parameter classification model. The preset engineering parameter classification model is a neural network model constructed by pre-training the classification of engineering parameters according to the TRIZ theory mentioned above. Therefore, based on the trained classification model, the result corresponding to the engineering parameter classification can be obtained based on the text attribute features.

[0535] Optionally, in some embodiments of the present invention, it may be possible to Fig.17 The device for constructing a classification model of text attribute features is used to train and construct the engineering parameter classification model, and the device for constructing a classification model of text attribute features mainly includes:

[0536] The sample acquisition module 171 is used to acquire engineering parameter corpus samples and text attribute feature samples, wherein the engineering parameter corpus samples include engineering parameter text attribute features and text attribute feature type tags. For example, for "increase weight, 1", where increase weight is an engineering parameter text attribute feature, 1 corresponding type tag indicates that increase weight belongs to the engineering parameter text attribute feature. If the tag is 0, it indicates that the corresponding phrase does not belong to the engineering parameter text attribute feature. In an optional embodiment, the text attribute feature samples described herein can be generated by identifying the text attribute feature described in any of the above embodiments;

[0537] The vector generation module 172 is used to generate an engineering parameter input vector according to the text attribute feature sample, and to generate an engineering parameter output vector according to the engineering parameter corpus sample. The input vector and output vector required for the training model are constructed based on the above samples.

[0538] The model building module 173 is used to train the SVM model according to the engineering parameter input vector and the engineering parameter output vector to build the engineering parameter classification model. In the embodiment of the present invention, each text attribute feature phrase can be classified into an engineering parameter text attribute feature using the SVM algorithm, and the Gaussian kernel is used to train the neural network model, thereby building the required engineering parameter classification model.

[0539] The embodiment of the present invention also provides a text structure analysis device, such as Fig.18 As shown, the text structure analysis device mainly includes:

[0540] The phrase set generation module 181 is used to identify phrases in each paragraph of the target text and form phrase sets according to each paragraph; the details can be found in the relevant description of step S131 of the above method embodiment, which will not be repeated here;

[0541] The structured set generation module 182 is used to perform structured analysis on each phrase set to generate a structured set; the details can be found in the relevant description of step S132 of the above method embodiment, which will not be repeated here;

[0542] The text recognition module 183 is used to recognize the text attribute features in the paragraphs of the structured set; the details can be found in the relevant description of step S133 of the above method embodiment, which will not be repeated here;

[0543] The text relationship building module 184 is used to establish the association relationship between the structured set and the phrase set according to the identified text attribute features; for details, please refer to the relevant description of step S134 of the above method embodiment, which will not be repeated here.

[0544] Through the above process, it is possible to recommend a structured technical solution for the analysis of text attribute features, or to predict the text attribute feature content of the structured technical solution). Moreover, it is also possible to combine certain restrictions to give more accurate results. For example, by combining efficacy and technical fields or keywords, a targeted technical solution can be given, or a more accurate efficacy prediction can be given for a certain structured technical solution and key components (multiple prediction results can be given, and the probability can be expressed as a percentage).

[0545] The embodiment of the present invention also provides a computer device, such as Fig.19 As shown, the computer device may include a processor 191 and a memory 192, wherein the processor 191 and the memory 192 may be connected via a bus or other means. Fig.19 The example of connecting through bus is taken in the following.

[0546] The processor 191 may be a central processing unit (CPU). The processor 191 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0547] The memory 192 is a non-transient computer-readable storage medium that can be used to store non-transient software programs, non-transient computer executable programs and modules, such as the program instructions / modules corresponding to the text attribute feature recognition method, or the text attribute feature classification method, or the text attribute feature classification model construction method in the embodiments of the present invention. The processor 191 executes various functional applications and data processing of the processor by running the non-transient software programs, instructions and modules stored in the memory 192, that is, to implement the text attribute feature recognition method, or the text attribute feature classification method, or the text attribute feature classification model construction method in the above method embodiments.

[0548] The memory 192 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and applications required for at least one function; the data storage area may store data created by the processor 191, etc. In addition, the memory 192 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 192 may optionally include a memory remotely arranged relative to the processor 191, and these remote memories may be connected to the processor 191 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0549] The one or more modules are stored in the memory 192 and when executed by the processor 191, perform the following steps: Figure 1-Figure 13 The text attribute feature recognition method in the illustrated embodiment, or the text attribute feature classification method, or the text attribute feature classification model construction method.

[0550] For details of the above computer equipment, please refer to Figures 1 to 13 The corresponding related descriptions and effects in the illustrated embodiments can be understood and will not be repeated here.

[0551] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.

[0552] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for identifying text attribute features, comprising: Acquiring a data structure in a target text, including: generating a grammatical structure according to sentences in the target text; generating a data structure according to node relationships in the grammatical structure; generating a first input vector according to the data structure; Determining the probability that each of the target texts contains attribute feature text according to the first input vector and a preset text attribute feature classification model; Identifying the text attribute features in the target text according to the probability, generating a grammatical structure according to the sentences in the target text, including: respectively identifying the words of each sentence in the target text and constructing a word directed graph; calculating the shortest path from the first node to the last node in the word directed graph as the word segmentation result of each sentence; constructing a word sequence according to the word segmentation result of each sentence; generating an input vector according to adjacent words in the word sequence; obtaining an output vector according to a preset neural network model and the input vector; calculating the cosine value of the angle between the input vector and the output vector; constructing a combined node from two adjacent words with the largest cosine value of the angle until a root node of the word sequence is generated; and determining the grammatical structure of the word sequence according to the combined node and the root node.

2. According to the text attribute feature recognition method of claim 1, the preset text attribute feature classification model is an efficacy sentence classification model, and the efficacy sentence classification model is constructed by the following process: Acquire a sample of efficacy statement, wherein the sample of efficacy statement includes a preset efficacy mark and a sample statement; Generate a first grammatical structure according to the efficacy sentence sample; generating a first feature list according to the node relationship in the first grammatical structure; Generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy mark and the sample sentence; The preset classification model is trained according to the first classification input vector and the first classification output vector to generate the efficacy statement classification model.

3. According to the text attribute feature recognition method of claim 2, the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process: Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase; generating a second grammatical structure according to the functional phrase sample; generating a second feature list according to the node relationship in the second grammatical structure; Generate a second data structure according to a preset efficacy phrase; generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase; The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

4. The method for recognizing text attribute features according to claim 3, wherein determining the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model comprises at least one of the following steps: Determining a first probability that each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model; Determining a second probability that a phrase in each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model; A third probability that a paragraph in each of the sentences contains an attribute feature is determined based on the first input vector and a preset text attribute feature classification model.

5. The method for recognizing text attribute features according to claim 2, wherein determining the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model comprises at least one of the following steps: Determining a first probability that each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model; Determining a second probability that a phrase in each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model; A third probability that a paragraph in each of the sentences contains an attribute feature is determined based on the first input vector and a preset text attribute feature classification model.

6. According to the text attribute feature recognition method of claim 1, the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process: Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase; generating a second grammatical structure according to the functional phrase sample; generating a second feature list according to the node relationship in the second grammatical structure; Generate a second data structure according to a preset efficacy phrase; generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase; The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

7. The method for recognizing text attribute features according to claim 6, wherein determining the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model comprises at least one of the following steps: Determining a first probability that each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model; Determining a second probability that a phrase in each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model; A third probability that a paragraph in each of the sentences contains an attribute feature is determined based on the first input vector and a preset text attribute feature classification model.

8. According to the text attribute feature recognition method of claim 1, the preset text attribute feature classification model is an efficacy sentence classification model, and the efficacy sentence classification model is constructed by the following process: Acquire a sample of efficacy statement, wherein the sample of efficacy statement includes a preset efficacy mark and a sample statement; Generate a first grammatical structure according to the efficacy sentence sample; generating a first feature list according to the node relationship in the first grammatical structure; Generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy mark and the sample sentence; The preset classification model is trained according to the first classification input vector and the first classification output vector to generate the efficacy statement classification model.

9. According to the text attribute feature recognition method of claim 8, the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process: Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase; generating a second grammatical structure according to the functional phrase sample; generating a second feature list according to the node relationship in the second grammatical structure; Generate a second data structure according to a preset efficacy phrase; generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase; The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

10. The method for recognizing text attribute features according to claim 9, wherein determining the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model comprises at least one of the following steps: Determining a first probability that each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model; Determining a second probability that a phrase in each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model; A third probability that a paragraph in each of the sentences contains an attribute feature is determined based on the first input vector and a preset text attribute feature classification model.

11. The method for recognizing text attribute features according to claim 1, wherein the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process: Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase; generating a second grammatical structure according to the functional phrase sample; generating a second feature list according to the node relationship in the second grammatical structure; Generate a second data structure according to a preset efficacy phrase; generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase; The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

12. The method for recognizing text attribute features according to claim 11, wherein determining the probability that each target text contains an attribute feature text according to the first input vector and a preset text attribute feature classification model comprises at least one of the following steps: Determining a first probability that each of the target texts contains an attribute feature according to the first input vector and a preset text attribute feature classification model; Determining a second probability that each phrase in the target text contains an attribute feature according to the first input vector and a preset text attribute feature classification model; A third probability that each paragraph in the target text contains an attribute feature is determined according to the first input vector and a preset text attribute feature classification model.

13. The method for recognizing text attribute features according to claim 1, wherein determining the probability that each target text contains an attribute feature text according to the first input vector and a preset text attribute feature classification model comprises at least one of the following steps: Determining a first probability that each of the target texts contains an attribute feature according to the first input vector and a preset text attribute feature classification model; Determining a second probability that each phrase in the target text contains an attribute feature according to the first input vector and a preset text attribute feature classification model; A third probability that each paragraph in the target text contains an attribute feature is determined according to the first input vector and a preset text attribute feature classification model.

14. A method for identifying text attribute features, comprising: Generate grammatical structures based on sentences in the target text; Generate a data structure according to the node relationship in the grammatical structure; generating a first input vector according to the data structure; Determining the probability that each sentence in the target text contains attribute feature text according to the first input vector and a preset text attribute feature classification model; Identify text attribute features in the target text according to the probability; The step of generating a grammatical structure according to sentences in the target text includes: Respectively identifying words of each sentence in the target text and constructing a word directed graph; Calculate the shortest path from the first node to the last node in the word directed graph as the word segmentation result of each of the sentences; Construct a word sequence based on the word segmentation results of each sentence; Generating an input vector based on adjacent words in the word sequence; Obtaining an output vector according to a preset neural network model and the input vector; Calculating the cosine value of the angle between the input vector and the output vector; Two adjacent words with the largest angle cosine value are used to construct a combined node until the root node of the word sequence is generated; The grammatical structure of the word sequence is determined according to the combination node and the root node.

15. The method for recognizing text attribute features according to claim 14, wherein the preset text attribute feature classification model is an efficacy sentence classification model, and the efficacy sentence classification model is constructed by the following process: Acquire a sample of efficacy statement, wherein the sample of efficacy statement includes a preset efficacy mark and a sample statement; Generate a first grammatical structure according to the efficacy sentence sample; generating a first feature list according to the node relationship in the first grammatical structure; Generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy mark and the sample sentence; The preset classification model is trained according to the first classification input vector and the first classification output vector to generate the efficacy statement classification model.

16. The text attribute feature recognition method according to claim 15, wherein the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process: Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase; generating a second grammatical structure according to the functional phrase sample; generating a second feature list according to the node relationship in the second grammatical structure; Generate a second data structure according to a preset efficacy phrase; generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase; The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

17. The text attribute feature recognition method according to claim 14, wherein the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process: Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase; generating a second grammatical structure according to the functional phrase sample; generating a second feature list according to the node relationship in the second grammatical structure; Generate a second data structure according to a preset efficacy phrase; generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase; The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

18. A method for identifying text attribute features, comprising: Generate grammatical structures based on sentences in the target text; Generate a data structure according to the node relationship in the grammatical structure; generating a first input vector according to the data structure; Determining the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model; Identify text attribute features in the target text according to the probability; The preset text attribute feature classification model is an efficacy sentence classification model, which is constructed by the following process: Acquire a sample of efficacy statement, wherein the sample of efficacy statement includes a preset efficacy mark and a sample statement; Generate a first grammatical structure according to the efficacy sentence sample; generating a first feature list according to the node relationship in the first grammatical structure; Generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy mark and the sample sentence; The preset classification model is trained according to the first classification input vector and the first classification output vector to generate the efficacy statement classification model.

19. The text attribute feature recognition method according to claim 18, wherein the preset text attribute feature classification model is a functional phrase classification model, and the functional phrase classification model is constructed by the following process: Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase; generating a second grammatical structure according to the functional phrase sample; generating a second feature list according to the node relationship in the second grammatical structure; Generate a second data structure according to a preset efficacy phrase; generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase; The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

20. A method for identifying text attribute features, comprising: Generate grammatical structures based on sentences in the target text; Generate a data structure according to the node relationship in the grammatical structure; generating a first input vector according to the data structure; Determining the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model; Identify text attribute features in the target text according to the probability; The preset text attribute feature classification model is a functional phrase classification model, which is constructed by the following process: Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase; generating a second grammatical structure according to the functional phrase sample; generating a second feature list according to the node relationship in the second grammatical structure; Generate a second data structure according to a preset efficacy phrase; generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase; The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

21. The method for recognizing text attribute features according to claim 15, 16, 17 or 19, characterized in that: Determining the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model comprises at least one of the following steps: Determining a first probability that each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model; Determining a second probability that a phrase in each of the sentences contains an attribute feature according to the first input vector and a preset text attribute feature classification model; A third probability that a paragraph in each of the sentences contains an attribute feature is determined based on the first input vector and a preset text attribute feature classification model.

22. A method for classifying text attribute features, comprising: Identify text attribute features in the target text according to sentences in the target text by using the text attribute feature recognition method according to any one of claims 1 to 21; Constructing a text attribute feature set according to the recognition result of the target text; Constructing a structure tree according to the text attribute feature set; Determine the class center point of each leaf node in the structure tree respectively, and form a class center set with multiple class center points; Generate a class center vector according to the plurality of class center points, and generate a leaf node vector according to the leaf nodes of the structure tree; The classification node of each leaf node is determined according to the class center vector and the leaf node vector classification node.

23. The method for classifying text attribute features according to claim 22, wherein the step of respectively determining the class center point of each leaf node in the structure tree comprises: Taking one of the leaf nodes in the structure tree as the central node respectively, and calculating the average distance from the leaf node in the structure tree to the central node; The leaf node with the farthest average distance is determined as the cluster center point.

24. The text attribute feature classification method according to claim 22, wherein the step of determining the classification node of each leaf node according to the class center vector and the leaf node vector classification node comprises: Calculating the Euclidean distance from the leaf node vector to the class center vector; For each leaf node, the class center point corresponding to the class center vector with the smallest Euclidean distance is used as the classification node of each leaf node.

25. A method for constructing a classification model of text attribute features, comprising: Acquire an engineering parameter corpus sample and a text attribute feature sample, wherein the engineering parameter corpus sample includes an engineering parameter text attribute feature and a text attribute feature type tag, and the text attribute feature sample is generated by identifying according to the text attribute feature identification method according to any one of claims 1 to 21; Generating an engineering parameter input vector according to the text attribute feature sample, and generating an engineering parameter output vector according to the engineering parameter corpus sample; The SVM model is trained according to the engineering parameter input vector and the engineering parameter output vector to construct an engineering parameter classification model.

26. A method for classifying text attribute features, comprising: Acquire a target text, and identify text attribute features of the target text by using the text attribute feature recognition method according to any one of claims 1 to 21; An engineering parameter classification result corresponding to the text attribute feature is generated according to the text attribute feature and a preset engineering parameter classification model.

27. According to the text attribute feature classification method of claim 26, the engineering parameter classification model is constructed according to the text attribute feature classification model construction method of claim 25.

28. A text structure analysis method comprising: Identify phrases in each paragraph of the target text and form phrase sets based on each paragraph; Perform structural analysis on each phrase set to generate a structured set; Identify text attribute features in paragraphs of the structured set according to the text attribute feature identification method according to any one of claims 1 to 21; An association relationship between the structured set and the phrase set is established according to the recognized text attribute features.

29. The text structure analysis method according to claim 28, If the paragraph of the structured set does not contain the text attribute feature, the first phrase set which is closest to the current phrase and located after the current phrase is selected and an association relationship is established with the structured set.

30. The text structure analysis method according to claim 29, If the first phrase set is not found, a second phrase set which is closest to the current phrase and located before the current phrase is selected and an association relationship is established with the structured set.

31. The text structure analysis method according to claim 30, If neither the first phrase set nor the second phrase set exists, the current phrase is ignored.

32. A device for identifying text attribute features, comprising: A data structure acquisition module is used to acquire the data structure of the target text; A first input vector generating module, used for generating a first input vector according to the data structure; A probability determination module, used to determine the probability that each of the target texts contains attribute feature text according to the first input vector and a preset text attribute feature classification model; A text recognition module, used for recognizing text attribute features in the target text according to the probability; The preset text attribute feature classification model is an efficacy sentence classification model, which is constructed by the following process: Acquire a sample of efficacy statement, wherein the sample of efficacy statement includes a preset efficacy mark and a sample statement; Generate a first grammatical structure according to the efficacy sentence sample; generating a first feature list according to the node relationship in the first grammatical structure; Generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy mark and the sample sentence; The preset classification model is trained according to the first classification input vector and the first classification output vector to generate the efficacy statement classification model.

33. A device for identifying text attribute features, comprising: A data structure acquisition module is used to acquire the data structure of the target text; A first input vector generating module, used for generating a first input vector according to the data structure; A probability determination module, used to determine the probability that the target text contains attribute feature text according to the first input vector and a preset text attribute feature classification model; A text recognition module, used for recognizing text attribute features in the target text according to the probability; The preset text attribute feature classification model is a functional phrase classification model, which is constructed by the following process: Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase; generating a second grammatical structure according to the functional phrase sample; generating a second feature list according to the node relationship in the second grammatical structure; Generate a second data structure according to a preset efficacy phrase; generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase; The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

34. A device for identifying text attribute features, comprising: A data structure acquisition module is used to acquire the data structure of the target text; A first input vector generating module, used for generating a first input vector according to the data structure; A probability determination module, used to determine the probability that the target text contains attribute feature text according to the first input vector and a preset text attribute feature classification model; A text recognition module, used for recognizing text attribute features in the target text according to the probability; Wherein, the data structure acquisition module is specifically used for: Generate grammatical structures based on sentences in the target text; Generate a data structure according to the node relationship in the grammatical structure; The step of generating a grammatical structure according to sentences in the target text includes: Respectively identifying words of each sentence in the target text and constructing a word directed graph; Calculate the shortest path from the first node to the last node in the word directed graph as the word segmentation result of each of the sentences; Construct a word sequence based on the word segmentation results of each sentence; Generating an input vector based on adjacent words in the word sequence; Obtaining an output vector according to a preset neural network model and the input vector; Calculating the cosine value of the angle between the input vector and the output vector; Two adjacent words with the largest angle cosine value are used to construct a combined node until the root node of the word sequence is generated; The grammatical structure of the word sequence is determined according to the combination node and the root node.

35. A device for identifying text attribute features, comprising: A grammatical structure generation module, used for generating a grammatical structure according to sentences in a target text; A data structure generation module, used to generate a data structure according to the node relationship in the grammatical structure; A first input vector generating module, used for generating a first input vector according to the data structure; A probability determination module, used to determine the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model; A text recognition module, used for recognizing text attribute features in the target text according to the probability; Wherein, the grammatical structure generation module is specifically used for: Respectively identifying words of each sentence in the target text and constructing a word directed graph; Calculate the shortest path from the first node to the last node in the word directed graph as the word segmentation result of each of the sentences; Construct a word sequence based on the word segmentation results of each sentence; Generating an input vector based on adjacent words in the word sequence; Obtaining an output vector according to a preset neural network model and the input vector; Calculating the cosine value of the angle between the input vector and the output vector; Construct a combined node from two adjacent words with the largest angle cosine value until a root node of the word sequence is generated; The grammatical structure of the word sequence is determined according to the combination node and the root node.

36. A text attribute feature recognition device, comprising: A grammatical structure generation module, used for generating a grammatical structure according to sentences in a target text; A data structure generation module, used to generate a data structure according to the node relationship in the grammatical structure; A first input vector generating module, used for generating a first input vector according to the data structure; A probability determination module, used to determine the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model; A text recognition module, used for recognizing text attribute features in the target text according to the probability; The preset text attribute feature classification model is an efficacy sentence classification model, which is constructed by the following process: Acquire a sample of efficacy statement, wherein the sample of efficacy statement includes a preset efficacy mark and a sample statement; Generate a first grammatical structure according to the efficacy sentence sample; generating a first feature list according to the node relationship in the first grammatical structure; Generating a first classification input vector according to the feature list, and generating a first classification output vector according to the preset efficacy mark and the sample sentence; The preset classification model is trained according to the first classification input vector and the first classification output vector to generate the efficacy statement classification model.

37. A device for identifying text attribute features, comprising: A grammatical structure generation module, used for generating a grammatical structure according to sentences in a target text; A data structure generation module, used to generate a data structure according to the node relationship in the grammatical structure; A first input vector generating module, used for generating a first input vector according to the data structure; A probability determination module, used to determine the probability that each of the sentences contains attribute feature text according to the first input vector and a preset text attribute feature classification model; A text recognition module, used for recognizing text attribute features in the target text according to the probability; The preset text attribute feature classification model is a functional phrase classification model, which is constructed by the following process: Acquire a sample of efficacy phrases, wherein the sample of efficacy phrases includes a preset efficacy mark and a sample phrase; generating a second grammatical structure according to the functional phrase sample; generating a second feature list according to the node relationship in the second grammatical structure; Generate a second data structure according to a preset efficacy phrase; generating a second classification input vector according to the second feature list and the second data structure, and generating a second classification output vector according to the preset efficacy mark and the sample phrase; The preset classification model is trained according to the second classification input vector and the second classification output vector to generate the preset efficacy phrase classification model.

38. A text attribute feature classification device, comprising: A text recognition module, configured to recognize text attribute features in a target text according to sentences in the target text by using a text attribute feature recognition method according to any one of claims 1 to 21; A text attribute feature set construction module, used to construct a text attribute feature set according to the recognition result of the target text; A structure tree construction module, used to construct a structure tree according to the text attribute feature set; A class center set construction module is used to determine the class center point of each leaf node in the structure tree respectively, and form a class center set with multiple class center points; A vector generation module, used to generate a class center vector according to the multiple class center points, and to generate a leaf node vector according to the leaf nodes of the structure tree; The classification node determination module is used to determine the classification node of each leaf node according to the class center vector and the leaf node vector classification node.

39. A device for constructing a classification model of text attribute features, comprising: A sample acquisition module, used for acquiring engineering parameter corpus samples and text attribute feature samples, wherein the engineering parameter corpus samples include engineering parameter text attribute features and text attribute feature type tags, and the text attribute feature samples are generated by identifying according to the text attribute feature identification method according to any one of claims 1 to 21; A vector generation module, used to generate an engineering parameter input vector according to the text attribute feature sample, and to generate an engineering parameter output vector according to the engineering parameter corpus sample; The model building module is used to train the SVM model according to the engineering parameter input vector and the engineering parameter output vector to build an engineering parameter classification model.

40. A device for classifying text attribute features, comprising: A text recognition module, used for acquiring a target text, and identifying a functional phrase of the target text by using a text attribute feature recognition method according to any one of claims 1 to 21; The text classification module is used to generate the engineering parameter classification result corresponding to the efficacy phrase according to the text attribute characteristics and a preset engineering parameter classification model.

41. A text structure analysis device, comprising: A phrase set generation module is used to identify phrases in each paragraph of the target text and form phrase sets according to each paragraph; A structured set generation module is used to perform structural analysis on each phrase set to generate a structured set; A text recognition module, configured to recognize text attribute features in paragraphs of the structured set according to the text attribute feature recognition method according to any one of claims 1 to 21; The text relationship building module is used to establish an association relationship between the structured set and the phrase set according to the identified text attribute features.

42. A computer device comprising: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method for recognizing text attribute features as described in any one of claims 1 to 21, or the method for classifying text attribute features as described in any one of claims 22 to 24, or the method for constructing a classification model of text attribute features as described in claim 25, or the method for classifying text attribute features as described in claim 26 or 27, or the method for analyzing text structure as described in any one of claims 28 to 31 by executing the computer instructions.

43. A computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, wherein the computer instructions are used to cause the computer to execute the method for identifying text attribute features as described in any one of claims 1-21, or to execute the method for classifying text attribute features as described in any one of claims 22-24, or to execute the method for constructing a classification model of text attribute features as described in claim 25, or to execute the method for classifying text attribute features as described in claim 26 or 27, or to execute the text structure analysis method as described in any one of claims 28-31.

Images