Key information extraction method and system of instruction type statement and electronic equipment

Abstract

The invention discloses a key information extraction method and system for an instruction type statement and electronic equipment. The method comprises the steps of obtaining a target instruction statement; analyzing and processing the target instruction statement to obtain a statement structure analysis graph; and extracting and integrating the statement structure analysis atlas, and extracting a target word in the statement structure analysis atlas as key information. According to the method, the statement structure analysis atlas is generated, the statement structure analysis atlas is extracted and integrated, and the target words in the statement structure analysis atlas are extracted to serve as key information, so that the sufficiency of statement analysis of the target instruction can be guaranteed, missing of statement analysis information is avoided, and a statement analysis result is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence, and particularly to a method, a system and an electronic device for extracting key information of directive statements. Background Art

[0002] With the development of agent interaction technology, for some artificial intelligence systems, such as intelligent question - answering systems, the interaction between users and agents is mainly in the form of natural language. During the interaction with agents, directive statements are a large category of interaction information. Various task requirements can be sent to agents through directive statements. Therefore, agents need to parse and understand various information in the statements, including but not limited to entity information, action information, and various associated constraint information. At the same time, specific instruction execution goals may also need to be generated.

[0003] In related prior arts, directive statements are mainly parsed through SDP (Semantic Dependency Parsing) or AMR (Abstract Meaning Representation), etc., and then rule extraction is usually combined with the parsing results to obtain the key information in the instructions. At the same time, there are also some neural - network - based information extraction methods that directly perform information extraction end - to - end and bypass explicit semantic parsing.

[0004] However, in the above - mentioned related prior arts, the method of solely using semantic dependency analysis cannot represent concepts that do not exist on the surface of some sentences; while the method of abstract semantic representation may lead to the loss of some important information. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art.

[0006] For this reason, one object of the present invention is to propose a method for extracting key information of directive statements. During the process of extracting key information of directive statements, by generating a statement structure parsing graph, extracting and integrating the statement structure parsing graph, and extracting the target words therein as key information, it can better serve downstream modules, avoid the problem of being unable to trace the source when errors occur, and ensure the sufficiency of the parsing of target directive statements, avoid the omission of statement parsing information, and make the parsing result more accurate.

[0007] For this reason, a second object of the present invention is to propose a system for extracting key information of directive statements.

[0008] For this reason, a third object of the present invention is to propose an electronic device.

[0009] To this end, the fourth object of the present invention is to propose a computer-readable storage medium.

[0010] To achieve the above object, an embodiment of the first aspect of the present invention discloses a method for extracting key information of an instruction-type statement, including the following steps: obtaining a target instruction statement; analyzing and processing the target instruction statement to obtain a statement structure parsing graph; extracting and integrating the statement structure parsing graph, and extracting target words therein as key information.

[0011] According to the method for extracting key information of an instruction-type statement according to an embodiment of the present invention, in the process of extracting key information of an instruction-type statement, by generating a statement structure parsing graph, extracting and integrating the statement structure parsing graph, and extracting target words therein as key information, it is possible to ensure the sufficiency of the parsing of the target instruction statement, avoid the omission of statement parsing information, and make the parsing result more accurate.

[0012] In addition, the method for extracting key information of an instruction-type statement according to the above embodiment of the present invention may further have the following additional technical features:

[0013] In some examples, the extracting and integrating the statement structure parsing graph, and extracting target words therein as key information includes: disassembling the statement structure parsing graph, and extracting the subject, predicate, object, and indirect object therein; matching the attributes required by the subject, predicate, object, and indirect object from the statement structure parsing graph, and using the subject, predicate, object, and indirect object with matched attributes as the key information.

[0014] In some examples, the attributes of the subject at least include: name; the attributes of the predicate at least include: at least one of tool, manner, and frequency; the attributes of the object and the indirect object at least include: at least one of number, individual quantifier, general attribute, location, and span.

[0015] In some examples, the subject and the object are entities, and the predicate is a verb or an adjective.

[0016] In some examples, the attributes of the object and the indirect object further include entity information, and the entity information at least includes: at least one of entity target, entity general attribute, entity description, entity image, entity data identification name, and entity label.

[0017] In some examples, the target instruction statement is parsed based on a semantic dependency analysis method to obtain a first semantic parsing graph; the target instruction statement is parsed based on an abstract semantic representation method to obtain a second semantic parsing graph; the first semantic parsing graph and the second semantic parsing graph are combined to obtain the statement structure parsing graph.

[0018] In some examples, combining the first semantic parsing graph and the second semantic parsing graph includes: using the second semantic parsing graph as a base graph; aligning nodes at corresponding positions in the base graph and the first semantic parsing graph to map the nodes in the first semantic parsing graph to corresponding nodes in the base graph; after node alignment, performing edge information optimization, where if there is edge information from the base graph, that edge information is preferentially used, otherwise, edge information from the first semantic parsing graph is used.

[0019] In some examples, when aligning nodes at corresponding positions in the base graph and the first semantic parsing graph, it further includes: if there are redundant nodes in the first semantic parsing graph, supplementing the redundant nodes in the first semantic parsing graph to the base graph.

[0020] In some examples, after extracting and integrating the statement structure parsing graph and extracting target words therein as key information, it further includes: generating a target interpretation graph based on the extracted target words, where the target interpretation graph includes an expected target state implicit in the target instruction statement.

[0021] In some examples, generating a target interpretation graph based on the extracted target words includes: when the target words form a subject-verb-object structure, if the object is attached with a semantic label related to orientation, determining that a positional relationship will be generated between the subject and the object under the corresponding action, and the structure of the generated target interpretation graph is subject - expected target state relation value - object, where the expected target state relation value is determined by the original word to which the orientation-related semantic label belongs.

[0022] In some examples, generating a target interpretation graph based on the extracted target words includes: when the target words form a subject-verb-object + indirect object structure, if the indirect object is attached with a semantic label related to orientation, determining that a positional relationship will be generated between the object and the indirect object under the corresponding action, and the structure of the generated target interpretation graph is object - expected target state relation value - indirect object, where the expected target state relation value is determined by the original word to which the orientation-related semantic label belongs.

[0023] To achieve the above object, an embodiment of the second aspect of the present invention discloses a key information extraction system for an instruction-type statement, including: an acquisition module for acquiring a target instruction statement; a first processing module for analyzing and processing the target instruction statement to obtain a statement structure parsing graph; a second processing module for extracting and integrating the statement structure parsing graph and extracting target words therein as key information.

[0024] According to the key information extraction system for directive statements of the embodiments of the present invention, during the process of extracting the key information of directive statements, by generating a statement structure analysis graph, extracting and integrating the statement structure analysis graph, and extracting the target words therein as key information, it can ensure the sufficiency of the analysis of the target directive statement, avoid the omission of statement analysis information, and make the analysis result more accurate.

[0025] To achieve the above object, the third aspect of the embodiments of the present invention discloses an electronic device, which includes: a processor, a memory, and a key information extraction program for directive statements stored on the memory and operable on the processor. When the key information extraction program for directive statements is executed by the processor, it implements the key information extraction method for directive statements as described in the first aspect of the embodiments of the present invention.

[0026] According to the electronic device of the embodiments of the present invention, during the process of extracting the key information of directive statements, by generating a statement structure analysis graph, extracting and integrating the statement structure analysis graph, and extracting the target words therein as key information, it can ensure the sufficiency of the analysis of the target directive statement, avoid the omission of statement analysis information, and make the analysis result more accurate.

[0027] To achieve the above object, the fourth aspect of the embodiments of the present invention discloses a computer-readable storage medium, on which a key information extraction program for directive statements is stored. When the key information extraction program for directive statements is executed by the processor, it implements the key information extraction method for directive statements as described in the first aspect of the embodiments of the present invention.

[0028] According to the computer-readable storage medium of the embodiments of the present invention, when the key information extraction program for directive statements stored thereon is executed by the processor, during the process of extracting the key information of directive statements, by generating a statement structure analysis graph, extracting and integrating the statement structure analysis graph, and extracting the target words therein as key information, it can ensure the sufficiency of the analysis of the target directive statement, avoid the omission of statement analysis information, and make the analysis result more accurate.

[0029] The additional aspects and advantages of the present invention will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The above and / or additional aspects and advantages of the present invention will become obvious and easy to understand from the description of the embodiments in conjunction with the following drawings, where:

[0031] Figure 1 is an exemplary diagram of the key information analysis of directive statements according to an embodiment of the present invention;

[0032] Figure 2 is an example diagram for parsing key information of directive statements according to another embodiment of the present invention;

[0033] Figure 3 is an example diagram for parsing key information of directive statements according to another embodiment of the present invention;

[0034] Figure 4 is a schematic flowchart of a method for extracting key information of directive statements according to an embodiment of the present invention;

[0035] Figure 5 is a schematic structural diagram of a system for extracting key information of directive statements according to an embodiment of the present invention.

[0036] Reference numerals:

[0037] Device for extracting key information of directive statements - 100; Acquisition module - 110; First processing module - 120; Second processing module - 130. Detailed implementation manners

[0038] The embodiments of the present invention will be described in detail below. The embodiments described with reference to the accompanying drawings are exemplary. The embodiments of the present invention will be described in detail below.

[0039] In the prior art, as Figure 1 shown, taking the directive statement "Gently place three red rags and two white iPhones on the table, and then turn off the computer" as an example. On the one hand, if the semantic dependency analysis method is used to parse this directive statement, although this method can directly obtain deep semantic information and thus mark the roles played by words. For example, it can identify the "rags and iPhones" led by "put", and the targets led by the two verbs "place on" and "turn off"; however, it cannot represent some concepts that do not exist on the surface of the sentence. For example, "iPhone" belongs to electronic devices and has no clear semantic connection with "rag", and the two are not related. On the other hand, as Figure 2 shown, if the abstract semantic representation method is used to parse this directive statement, although this method can freely introduce concepts that do not exist on the surface of the sentence and has stronger expression ability. For example, it can represent the parallel relationship between "three red rags" and "two white iPhones" and the time sequence relationship of first "place on the table" and then "turn off the computer"; however, it has corpus limitations and model limitations, which will lead to the loss of some important information, such as the "on" in "on the table" in the statement.

[0040] Based on this, the present invention proposes a method for extracting key information from imperative statements. By fusing the parsing results of imperative statements based on semantic dependency analysis method and the parsing results of imperative statements based on abstract semantic representation method, the sufficiency of imperative statement parsing can be ensured, the omission of statement parsing information can be avoided, and the statement parsing result can be made more accurate. For example, taking the imperative statement "Gently place three red rags and two white iPhones on the table, and then turn off the computer" as an example, the structure diagram obtained based on the abstract semantic representation method, that is Figure 2 , and the structure diagram obtained based on the semantic dependency analysis method, that is Figure 1 , are combined to obtain a new structure diagram, that is Figure 3 . As shown in Figure 3 , the new structure diagram contains both the semantic dependency relationships of all words obtained by the semantic dependency analysis method and the temporal relationships between words obtained by the abstract semantic representation method. For example, the targets guided by the two verbs "place" and "turn off" and their overall temporal relationships further enrich the semantic expression of the imperative statement and improve the accuracy and flexibility of the parsing of the imperative statement.

[0041] Next, refer to Figures 4 - 5 to describe the method, system, electronic device and computer-readable storage medium for extracting key information from imperative statements according to the embodiments of the present invention.

[0042] Figure 4 is a schematic flowchart of a method for extracting key information from imperative statements according to an embodiment of the present invention. As shown in Figure 5 , the method for extracting key information from imperative statements includes the following steps:

[0043] Step S1: Obtain the target imperative statement.

[0044] Specifically, in the process of extracting key information from imperative statements, first, the target imperative statement can be obtained, including but not limited to obtaining the target imperative statement from written texts such as written texts, emails, instant messages, etc., or screening out the target imperative statement from large-scale text data by using natural language processing tools and technologies such as semantic analysis and keyword extraction, or guiding the user to input the target imperative statement through an interactive interface or dialogue system. For example, obtaining the instruction input by the user in the input box or dialog box in the application.

[0045] Step S2: Analyze and process the target imperative statement to obtain a statement structure parsing map.

[0046] Specifically, after obtaining the target instruction statement, it is possible to analyze and process the target instruction statement, including the internal logical relationship and semantic information of the instruction statement, in order to obtain a statement structure parsing graph. Specifically, it is possible to perform syntactic and semantic analysis on the target instruction statement, decompose the text into linguistic components, identify semantic information such as entities, actions, and objects therein, and infer the semantic logical relationships between them, including subject-predicate-object relationships, modification relationships, and subordinate relationships, etc., thereby constructing a statement structure parsing graph to represent the syntactic structure and semantic relationships of the instruction statement.

[0047] Step S3: Extract and integrate the statement structure parsing graph, and extract the target words therein as key information.

[0048] Specifically, after obtaining the statement structure parsing graph, it is possible to identify information related to the instruction from the statement structure parsing graph, such as the subject, predicate, and object, etc., and integrate the information to extract the target words therein as key information. It can be understood that the key information can be verbs with behavioral properties and related noun phrases or modifying words, etc. For example, two actions "put, turn off" can be extracted from "Gently put three red rags and two white iPhones on the table, and then turn off the computer".

[0049] Therefore, in the process of extracting the key information of the above-mentioned instruction-type statement, by generating a statement structure parsing graph, extracting and integrating the statement structure parsing graph, and extracting the target words therein as key information, the sufficiency of the target instruction statement parsing is ensured, the omission of statement parsing information is avoided, and the parsing result is made more accurate.

[0050] In an embodiment of the present invention, extracting and integrating the statement structure parsing graph and extracting the target words therein as key information includes: disassembling the statement structure parsing graph, and extracting the subject, predicate, object, and indirect object therein; matching the attributes required by the subject, predicate, object, and indirect object respectively from the statement structure parsing graph, and taking the subject, predicate, object, and indirect object that have matched the attributes as key information.

[0051] Specifically, in the process of extracting and integrating the sentence structure parsing graph and extracting the target words therein as key information, the sentence structure parsing graph can be disassembled into different components, and the subject, predicate, object, and indirect object can be extracted therefrom; then, the attributes required for the subject, predicate, object, and indirect object are matched from the sentence structure parsing graph, including but not limited to semantic information, grammatical information, or other attributes related to the components, such as entity names, attribute values, etc., and the attributes can be flexibly configured; finally, the subject, predicate, object, and indirect object with matched attributes are used as key information to indicate the object or target in the instruction.

[0052] In an embodiment of the present invention, the attributes of the subject at least include: name; the attributes of the predicate at least include at least one of: tool, manner, frequency; the attributes of the object and indirect object at least include at least one of: number, individual quantifier, general attribute, location, span.

[0053] Specifically, among the extracted key information, the attributes of the subject at least include name to represent the object to be operated in the instruction; the attributes of the predicate at least include at least one of tool, manner, frequency to represent the specific details and requirements of the behavior or operation in the instruction, where the tool refers to the tool or device used for a certain behavior or operation; the manner refers to the way or method of a certain behavior or operation; the frequency refers to the frequency or number of times of a certain behavior or operation; the attributes of the object and indirect object at least include at least one of number, individual quantifier, general attribute, location, span to represent the specific information of the object and related attributes involved in the behavior or operation in the instruction, where the number is used to indicate the specific quantity involved in the behavior or operation, such as "one, two, three", etc.; the individual quantifier is used to indicate the individual quantity or type involved in the behavior or operation, such as "piece, branch, sheet", etc.; the general attribute is used to indicate the general attribute or feature involved in the behavior or operation, such as "red, black, blue", etc.; the location is used to indicate the place or position where the behavior or operation occurs; the span is used to indicate the time range or object range involved in the behavior or operation.

[0054] In an embodiment of the present invention, the subject and object are entities, and the predicate is a verb or an adjective.

[0055] Specifically, among the extracted key information, the subject and object are entities, that is, specific people or things, etc., such as "computer", "rag", "apple", etc.; the predicate is a verb or an adjective, where the verb is used to describe the behavior or action performed by the subject, such as "put", "turn off", etc.; the adjective is used to describe the state or feature of the subject or object, such as "red", "white", etc.

[0056] In one embodiment of the present invention, the attributes of the object and the indirect object further include entity information, which at least includes at least one of: entity target, entity general attributes, entity description, entity image, entity data identification name, and entity label.

[0057] Specifically, among the extracted key information, the attributes of the object and the indirect object further include entity information, which at least includes at least one of entity target, entity general attributes, entity description, entity image, entity data identification name, and entity label. Specifically, the entity target represents the specific entity or object referred to by the object or the indirect object, the entity general attributes represent the general attributes or characteristics of the entity referred to by the object or the indirect object, and the attributes of the entity information can be flexibly configured. The entity description represents the text description or explanation of the object or the indirect object, the entity image represents the image or picture of the entity referred to by the object or the indirect object, the entity data identification name represents the data identification or name of the entity referred to by the object or the indirect object, and the entity label represents the label or category of the entity referred to by the object or the indirect object.

[0058] In a specific embodiment, if the entity target is "rag", its general attributes can be "encyclopedia", the description can be "rag, pinyin mābù, a cloth used to wipe utensils; a kind of inner court official's clothing accessory in the Ming Dynasty", the data identification name can be "Chinese name, rag; English name, rag; pinyin, mābù"; phonetic notation, ㄇㄚㄅㄨˋ1.; uses, wipe tables, chairs, benches; materials, linen, silk, pure cotton, nylon, fiber", and the label can be "household items, life".

[0059] In a specific embodiment, a specific example of the extraction rule is as follows:

[0060]

[0061]

[0062]

[0063]

[0064]

[0065]

[0066] In the above example, the attributes of the entity information involved can be flexibly configured. Among them, Action represents an action, subject represents the subject, verb represents the verb, which belongs to the predicate, instrument represents the tool, manner represents the manner, frequency represents the frequency, objects represents the object, quant represents the number, cunit represents the individual quantifier, domain represents the general attribute, location represents the place, range represents the span, entity_info represents the entity information, entity represents the entity target, the general attribute represents the general entity attribute, desc represents the entity description, image represents the entity image, avp represents the entity data identification name, and tag represents the entity tag.

[0067] In an embodiment of the present invention, the target instruction statement is analyzed and processed to obtain a statement structure parsing graph, including: parsing the target instruction statement based on the semantic dependency analysis method to obtain a first semantic parsing graph; parsing the target instruction statement based on the abstract semantic representation method to obtain a second semantic parsing graph; combining the first semantic parsing graph and the second semantic parsing graph to obtain a statement structure parsing graph.

[0068] Specifically, in the process of analyzing and processing the target instruction statement, the target instruction statement can be parsed based on the semantic dependency analysis method to identify the subject, predicate, object, etc. in the statement and establish the dependency relationship between the words in the statement to obtain a first semantic parsing graph. Among them, the target instruction statement can be regarded as a directed graph, each word is a node in the graph, and each dependency relationship is an edge in the graph; at the same time, the target instruction statement can be parsed based on the abstract semantic representation method to convert the words in the statement into an abstract semantic representation and introduce concepts that do not exist on the surface of the statement to obtain a second semantic parsing graph. Among them, the target instruction statement can be regarded as a directed graph, each word is a node in the graph, and each semantic relationship is an edge in the graph; further, the first semantic parsing graph obtained based on the semantic dependency analysis method and the second semantic parsing graph obtained based on the abstract semantic representation method are combined to combine the advantages of the two methods to obtain a more comprehensive and accurate statement structure parsing graph.

[0069] In an embodiment of the present invention, combining the first semantic parsing graph and the second semantic parsing graph includes: using the second semantic parsing graph as the basic graph; aligning the nodes at the corresponding positions in the basic graph and the first semantic parsing graph to map the nodes in the first semantic parsing graph to the corresponding nodes in the basic graph; after the node alignment, perform edge information optimization. Among them, if there is edge information from the basic graph, give priority to using this edge information; otherwise, use the edge information from the first semantic parsing graph.

[0070] Specifically, since there may be problems of adding or ignoring some graph nodes in the second semantic parsing graph obtained based on the abstract semantic representation method, the second semantic parsing graph can be uniformly selected as the base graph, and supplementary operations can be performed on the base graph to ensure that information is not omitted. Specifically, by comparing the nodes at the corresponding positions in the base graph spectrum and the first semantic parsing graph spectrum, the node alignment between the base graph spectrum and the first semantic parsing graph spectrum can be achieved, so as to map the nodes in the first semantic parsing graph spectrum to the corresponding nodes in the base graph spectrum. For example, if the position of "red" in the command statement is [3, 5) (that is, the substring composed of the 3rd to 4th characters of the original string), then the nodes with the same position number [3, 5) in the base graph spectrum and the first semantic parsing graph spectrum both correspond to the same "red". Further, after the node alignment, there may be edge information from both the base graph spectrum and the first semantic parsing graph spectrum between two nodes. It is necessary to select the better edge information from the base graph spectrum and the first semantic parsing graph spectrum. Since the base graph spectrum is obtained based on the abstract semantic representation method, the semantic information attached to its edge information is more abundant and beneficial to downstream tasks. Therefore, if there is edge information from the base graph spectrum, this edge information is preferentially used; otherwise, the edge information from the first semantic parsing graph spectrum is used.

[0071] In an embodiment of the present invention, when aligning the nodes at the corresponding positions in the base graph spectrum and the first semantic parsing graph spectrum, it further includes: if there are redundant nodes in the first semantic parsing graph spectrum, the redundant nodes in the first semantic parsing graph spectrum are supplemented to the base graph spectrum.

[0072] Specifically, when aligning the nodes at the corresponding positions in the base graph spectrum and the first semantic parsing graph spectrum, if there are redundant nodes in the first semantic parsing graph spectrum, the redundant nodes can be supplemented to the base graph spectrum. The supplementation methods include but are not limited to corresponding to the redundant nodes by adjusting or adding nodes and edges in the base graph spectrum. It can be understood that since there may be problems of adding or ignoring some graph nodes in the second semantic parsing graph obtained based on the abstract semantic representation method, after the node alignment between the base graph spectrum and the first semantic parsing graph spectrum, when there are still redundant nodes in the first semantic parsing graph spectrum, the redundant nodes can be supplemented to the base graph spectrum.

[0073] In an embodiment of the present invention, after extracting and integrating the statement structure parsing graph spectrum and extracting the target words therein as key information, it further includes: generating a target interpretation graph according to the extracted target words, and the target interpretation graph includes the expected target state implied in the target instruction statement.

[0074] Specifically, after extracting and integrating the parsing graph of the statement structure and extracting the target words therein as key information, a target interpretation graph can be generated based on the extracted target words to clarify the target state expected by the instruction statement. Specifically, the target interpretation graph can include nodes and edges. The nodes represent the target words, and the edges represent the semantic connections or structural relationships between the nodes, that is, the nodes and edges of the target interpretation graph can represent the implicit target state in the instruction statement.

[0075] In an embodiment of the present invention, generating a target interpretation graph according to the extracted target words includes: when the target words form a subject-predicate-object structure, if the object is attached with a semantic label related to orientation, it is determined that a positional relationship will be generated between the subject and the object under the corresponding action, and the structure of the generated target interpretation graph is subject-expected target state relationship value-object, where the expected target state relationship value is determined by the original word to which the semantic label related to orientation belongs.

[0076] Specifically, when the target words form a subject-predicate-object structure, if the object is attached with a semantic label related to orientation, such as "left" in "Please walk to the left of the table", it indicates that a positional relationship will be generated between the subject and the object under the corresponding action, and the structure of the generated target interpretation graph is subject-expected target state relationship value-object, such as "you-on the left of-table". Among them, the expected target state relationship value is determined by the original word to which the semantic label related to orientation belongs. For example, if the semantic label related to orientation of the object is "above", the expected target state relationship value can be "above".

[0077] In an embodiment of the present invention, generating a target interpretation graph according to the extracted target words includes: when the target words form a subject-predicate-object + indirect object structure, if the indirect object is attached with a semantic label related to orientation, it is determined that a positional relationship will be generated between the object and the indirect object under the corresponding action, and the structure of the generated target interpretation graph is object-expected target state relationship value-indirect object, where the expected target state relationship value is determined by the original word to which the semantic label related to orientation belongs.

[0078] Specifically, when the target words form a subject-predicate-object and indirect object structure, if the indirect object is attached with a semantic label related to orientation, such as "on" in "Put the phone on the table", it indicates that a positional relationship will be generated between the object and the indirect object under the corresponding action, and the structure of the generated target interpretation graph is object-expected target state relationship value-indirect object, such as "phone-on...-table". Among them, the expected target state relationship value is determined by the original word to which the semantic label related to orientation belongs. For example, if the semantic label related to orientation of the indirect object is "below", the expected target state relationship value can be "below".

[0079] In summary, in the method for extracting key information of the above-mentioned directive statement, during the process of extracting key information of the directive statement, by generating a parsing graph of the statement structure, extracting and integrating the parsing graph of the statement structure, and extracting the target words therein as key information, the sufficiency of the parsing of the target directive statement can be ensured, and the parsing result can be made more accurate. Further, through the extracted key information, a target interpretation graph can be generated as the target basis for downstream execution, better serving the downstream module and avoiding the problem of being unable to trace the source when an error occurs.

[0080] A further embodiment of the present invention also proposes a key information extraction system 100 for directive statements.

[0081] Figure 5 It is a schematic structural diagram of a key information extraction system 100 for directive statements according to an embodiment of the present invention. As Figure 5 shown, the key information extraction system 100 for directive statements includes: an acquisition module 110, a first processing module 120, and a second processing module 130.

[0082] Specifically, the acquisition module 110 is used to acquire a target directive statement.

[0083] The first processing module 120 is used to analyze and process the target directive statement to obtain a parsing graph of the statement structure.

[0084] The second processing module 130 is used to extract and integrate the parsing graph of the statement structure, and extract the target words therein as key information.

[0085] In an embodiment of the present invention, the second processing module 130 extracts and integrates the parsing graph of the statement structure, and extracts the target words therein as key information, including: disassembling the parsing graph of the statement structure, and extracting the subject, predicate, object, and indirect object therein; matching the attributes required by the subject, predicate, object, and indirect object from the parsing graph of the statement structure, and taking the subject, predicate, object, and indirect object that have matched the attributes as key information.

[0086] In an embodiment of the present invention, the attributes of the subject at least include: name; the attributes of the predicate at least include: at least one of tool, manner, and frequency; the attributes of the object and indirect object at least include: at least one of number, individual quantifier, general attribute, location, and span.

[0087] In an embodiment of the present invention, the subject and object are entities, and the predicate is a verb or an adjective.

[0088] In one embodiment of the present invention, the attributes of the object and the indirect object further include entity information, and the entity information includes at least one of: entity target, entity general attributes, entity description, entity image, entity data identification name, and entity label.

[0089] In one embodiment of the present invention, the first processing module 120 analyzes and processes the target instruction statement to obtain a statement structure parsing graph, including: parsing the target instruction statement based on the semantic dependency analysis method to obtain a first semantic parsing graph; parsing the target instruction statement based on the abstract semantic representation method to obtain a second semantic parsing graph; and combining the first semantic parsing graph and the second semantic parsing graph to obtain a statement structure parsing graph.

[0090] In one embodiment of the present invention, the first processing module 120 combines the first semantic parsing graph and the second semantic parsing graph, including: using the second semantic parsing graph as the basic graph; aligning the nodes at the corresponding positions in the basic graph and the first semantic parsing graph to map the nodes in the first semantic parsing graph to the corresponding nodes in the basic graph; and after the node alignment, performing edge information optimization, where if there is edge information from the basic graph, the edge information from the basic graph is preferentially used, otherwise, the edge information from the first semantic parsing graph is used.

[0091] In one embodiment of the present invention, when the first processing module 120 aligns the nodes at the corresponding positions in the basic graph and the first semantic parsing graph, it further includes: if there are redundant nodes in the first semantic parsing graph, supplementing the redundant nodes in the first semantic parsing graph to the basic graph.

[0092] In one embodiment of the present invention, after the second processing module 130 extracts and integrates the statement structure parsing graph and extracts the target words therein as key information, it is further used for: generating a target interpretation graph according to the extracted target words, and the target interpretation graph includes the expected target state implicit in the target instruction statement.

[0093] In one embodiment of the present invention, the second processing module 130 generates a target interpretation graph according to the extracted target words, including: when the target words form a subject-verb-object structure, if the object is attached with a semantic label related to orientation, it is determined that there will be a position relationship between the subject and the object under the corresponding action, and the structure of the generated target interpretation graph is subject-expected target state relationship value-object, where the expected target state relationship value is determined by the original word to which the semantic label related to orientation belongs.

[0094] In an embodiment of the present invention, the second processing module 130 generates a target interpretation graph according to the extracted target words, including: when the target words form a subject-predicate-object + indirect object structure, if the indirect object is attached with a semantic label related to orientation, it is determined that a positional relationship will be generated between the object and the indirect object under the corresponding action, and the structure of the generated target interpretation graph is object - expected target state relationship value - indirect object, where the expected target state relationship value is determined by the original word to which the semantic label related to orientation belongs.

[0095] According to the key information extraction system 100 for imperative statements of the embodiments of the present invention, in the process of extracting the key information of imperative statements, by generating a statement structure parsing graph, extracting and integrating the statement structure parsing graph, and extracting the target words therein as key information, the sufficiency of the parsing of the target instruction statement can be ensured, and the parsing result can be made more accurate. Further, through the extracted key information, a target interpretation graph can be generated as the target basis for downstream execution, better serving the downstream module and avoiding the problem of being unable to trace back when an error occurs.

[0096] A further embodiment of the present invention provides an electronic device.

[0097] In some embodiments, the electronic device includes: a processor, a memory, and a key information extraction program for imperative statements stored on the memory and executable on the processor. When the key information extraction program for imperative statements is executed by the processor, it implements the key information extraction method for imperative statements as described in the first aspect embodiments above.

[0098] According to the electronic device of the embodiments of the present invention, in the process of extracting the key information of imperative statements, by generating a statement structure parsing graph, extracting and integrating the statement structure parsing graph, and extracting the target words therein as key information, the sufficiency of the parsing of the target instruction statement can be ensured, and the parsing result can be made more accurate. Further, through the extracted key information, a target interpretation graph can be generated as the target basis for downstream execution, better serving the downstream module and avoiding the problem of being unable to trace back when an error occurs.

[0099] A further embodiment of the present invention also discloses a computer-readable storage medium, on which a key information extraction program for imperative statements is stored. When the key information extraction program for imperative statements is executed by the processor, it implements the key information extraction method for imperative statements as described in the first aspect embodiments above.

[0100] When the key information extraction program of the instruction-type statement stored on the computer-readable storage medium according to the embodiment of the present invention is executed by a processor, if in the process of extracting the key information of the instruction-type statement, by generating a statement structure analysis graph, extracting and integrating the statement structure analysis graph, and extracting the target words therein as the key information, the sufficiency of the target instruction statement analysis can be ensured, and the analysis result can be made more accurate. Further, through the extracted key information, a target interpretation graph can be generated as the target basis for downstream execution, better serving the downstream module and avoiding the problem that it is impossible to trace the source when an error occurs.

[0101] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example.

[0102] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and purposes of the present invention, and the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for extracting key information of instruction statements, characterized in that Including: Obtain a target instruction statement; Analyze and process the target instruction statement to obtain a statement structure parsing graph; Extract and integrate the statement structure parsing graph, and extract target words therein as key information.

2. The key information extraction method for directive statements according to claim 1, characterized in that, The extracting and integrating the statement structure parsing graph and extracting target words therein as key information includes: Decompose the statement structure parsing graph, and extract the subject, predicate, object, and indirect object therein; Match the attributes required by the subject, predicate, object, and indirect object respectively from the statement structure parsing graph, and use the subject, predicate, object, and indirect object with matched attributes as the key information.

3. The method for extracting key information of an instruction-type statement according to claim 2, wherein The attributes of the subject at least include: name; The attributes of the predicate at least include at least one of: tool, manner, frequency; The attributes of the object and the indirect object at least include at least one of: number, individual quantifier, general attribute, location, span.

4. The method for extracting key information of directive statements according to claim 2, characterized in that, The subject and the object are entities, and the predicate is a verb or an adjective.

5. The key information extraction method for directive statements according to claim 3, characterized in that The attributes of the object and the indirect object further include entity information, and the entity information at least includes at least one of: entity target, entity general attribute, entity description, entity image, entity data identification name, entity label.

6. The key information extraction method for directive statements according to claim 1, characterized in that The analyzing and processing the target instruction statement to obtain a statement structure parsing graph includes: Parse the target instruction statement based on a semantic dependency analysis method to obtain a first semantic parsing graph; Parse the target instruction statement based on an abstract semantic representation method to obtain a second semantic parsing graph; Combine the first semantic parsing graph and the second semantic parsing graph to obtain the statement structure parsing graph.

7. The method for extracting key information of an instruction statement according to claim 6, characterized in that, The combining the first semantic parsing graph and the second semantic parsing graph includes: Use the second semantic parsing graph as the basic graph; Align the nodes at corresponding positions in the basic graph and the first semantic parsing graph, so as to correspond the nodes in the first semantic parsing graph to the corresponding nodes in the basic graph; After the node alignment, perform edge information optimization. Among them, if there is edge information from the basic graph, give priority to using this edge information; otherwise, use the edge information from the first semantic parsing graph.

8. The key information extraction method for directive statements according to claim 7, characterized in that When aligning the nodes at corresponding positions in the basic graph and the first semantic parsing graph, it further includes: If there are redundant nodes in the first semantic parsing graph, supplement the redundant nodes in the first semantic parsing graph to the basic graph.

9. The method for extracting key information of an instruction statement according to claim 2, wherein After extracting and integrating the statement structure parsing graph and extracting target words therein as key information, it further includes: Generate a target interpretation graph according to the extracted target words, and the target interpretation graph includes the expected target state implied in the target instruction statement.

10. The key information extraction method for directive statements according to claim 9, characterized in that, The generating a target interpretation graph according to the extracted target words includes: When the target words form a subject-predicate-object structure, if the object is attached with a semantic tag related to orientation, determine the positional relationship that will be generated between the subject and the object under the corresponding action, and the structure of the generated target interpretation graph is subject-expected target state relationship value-object, where the expected target state relationship value is determined by the original word to which the semantic tag related to orientation belongs.

11. The key information extraction method for directive statements according to claim 9, characterized in that, The generating the target interpretation graph according to the extracted target words includes: When the target words form a subject-predicate-object + indirect object structure, if the indirect object is attached with a semantic tag related to orientation, determine the positional relationship that will be generated between the object and the indirect object under the corresponding action, and the structure of the generated target interpretation graph is object-expected target state relationship value-indirect object, where the expected target state relationship value is determined by the original word to which the semantic tag related to orientation belongs.

12. A key information extraction system for instruction-type statements, characterized in that, including: an acquisition module, configured to acquire a target instruction statement; a first processing module, configured to perform analysis and processing on the target instruction statement to obtain a statement structure parsing graph; a second processing module, configured to extract and integrate the statement structure parsing graph, and extract the target words therein as key information.

13. An electronic device, characterized in that, including: a processor, a memory, and a key information extraction program for instruction-type statements stored on the memory and executable on the processor, where when the key information extraction program for instruction-type statements is executed by the processor, it implements the key information extraction method for instruction-type statements according to any one of claims 1-11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a key information extraction program for instruction-type statements, where when the key information extraction program for instruction-type statements is executed by the processor, it implements the key information extraction method for instruction-type statements according to any one of claims 1-11.

Images