Emergency field intention information extraction method based on dependency grammar analysis

Abstract

The invention discloses an emergency field intention information extraction method based on dependency grammar analysis. The method comprises the following steps: acquiring a semantic complete text of a user; performing intention classification on the semantic complete text by utilizing a text classification model to obtain intentions; obtaining a to-be-filled slot position corresponding to the intention according to the intention; performing dependency grammar analysis on the semantic complete text to obtain a dependency grammar analysis result; splicing the dependency grammar analysis result through an engineering means to obtain a dependency grammar transfer result; performing similarity retrieval in an example database by using a dependency grammar transfer result to obtain an approximate example; obtaining a cue word template; wherein the cue word template comprises example step construction; the example step constructs a dependency grammar transfer result comprising an approximate example and the approximate example; and inputting the slot position to be filled, the cue word template and the semantic complete text into the large language model to obtain an intention information extraction result. According to the invention, intentions of users in different emergency service scenes can be accurately extracted.

Description

Technical Field

[0001] The present invention relates to the technical field of intention information extraction, and more specifically to an emergency field intention information extraction method based on dependency grammar analysis. Background Art

[0002] In the development of artificial intelligence and machine learning, intelligent conversational robots have become a key application scenario. These robots provide users with accurate information and services by understanding and responding to user intentions. In this process, intent recognition algorithms play a key role. Currently, the most commonly used intent recognition algorithm is the BERT-based natural language processing (NLP) algorithm. This algorithm uses a deep bidirectional model, and the pre-trained language representation can capture the contextual information of the sentence, so that it can understand the user's intention.

[0003] However, although NLP algorithms such as BERT have achieved remarkable results in many fields, their effects are limited in command center scene instructions in the emergency field. The language environment in the emergency field has its own particularity and often contains a large number of professional words and terms. The meanings of these professional words and terms may be different from those in the conventional context. Therefore, traditional NLP algorithms often find it difficult to cover these situations. This leads to the fact that traditional NLP algorithms often fail to achieve the expected results in intent recognition in the emergency field.

[0004] In addition, the emergency field covers many different business scopes, such as fire, earthquake, flood, disease outbreak, etc. The language environment of these business scopes is different, and there are high requirements for the business scenario migration capability of the intent recognition algorithm. An ideal emergency field intent recognition algorithm should be able to accurately understand and identify the user's intention in different emergency business scenarios, so as to provide users with accurate services.

[0005] Therefore, how to provide a method for extracting intention information in the emergency field, which can accurately extract the intention of users in different emergency business scenarios and thus provide users with accurate services is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0006] In view of this, the purpose of the present invention is to provide a method for extracting emergency domain intention information based on dependency grammar analysis.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] The first aspect provides a method for extracting emergency domain intention information based on dependency grammar analysis, comprising the following steps:

[0009] S1: Obtain the user's semantically complete text;

[0010] S2: using a text classification model to classify the semantically complete text to obtain the intent of the semantically complete text;

[0011] According to the intention of the semantically complete text, a slot to be filled corresponding to the intention is obtained;

[0012] Performing dependency grammar analysis on the semantically complete text to obtain a dependency grammar analysis result of the semantically complete text;

[0013] The dependency grammar analysis results of the semantically complete text are assembled by engineering means to obtain the dependency grammar paraphrase results of the semantically complete text;

[0014] S3: using the dependency grammar paraphrase result of the semantically complete text to perform similarity search in the example database to obtain an approximate example of the semantically complete text;

[0015] S4: Acquire a prompt word template; wherein the prompt word template includes an example step construction; the example step construction includes the approximate example and the dependency grammar paraphrase result of the approximate example;

[0016] S5: Input the slot to be filled, the prompt word template and the semantically complete text into the large language model to obtain the intention information extraction result of the user's current input text.

[0017] Preferably, S1 specifically includes:

[0018] Get the user's current input text;

[0019] The user's current input text and the user's historical input text are input into the contextual multi-round understanding module to obtain the user's semantically complete text.

[0020] Preferably, S2 specifically includes:

[0021] S31: preprocessing the semantically complete text;

[0022] S32: converting the preprocessed text into numerical features;

[0023] S33: Input the numerical features into a text classification model for classification to obtain a predefined intent corresponding to the semantically complete text; wherein the predefined intent corresponding to the semantically complete text is the intent of the semantically complete text.

[0024] Preferably, the preprocessing includes word segmentation and removal of stop words;

[0025] Convert the preprocessed text into numerical features using bag-of-words (BOW), TF-IDF, or word embedding;

[0026] The text classification model is Naive Bayes or Support Vector Machine.

[0027] Preferably, similarity retrieval is performed based on the number and type of dependency relations in the dependency grammar paraphrase result of the semantically complete text.

[0028] Preferably, the prompt word template also includes model role definition, intent information extraction slot definition, intent information extraction results of approximate examples and task requirements.

[0029] Preferably, the task requires that the final output intent information extraction result be in the format of json.

[0030] The second aspect provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for extracting emergency domain intention information based on dependency grammar analysis as described in any one of the above items is implemented.

[0031] The third aspect provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the emergency domain intention information extraction method based on dependency grammar analysis as described in any one of the above items.

[0032] A fourth aspect provides a computer program product, comprising a computer program, which, when executed by a processor, implements the emergency domain intention information extraction method based on dependency grammar analysis as described in any one of the above items.

[0033] It can be seen from the above technical solution that, compared with the prior art, the present invention discloses a method for extracting intention information in the emergency field based on dependency grammar analysis. The present invention can accurately extract the intentions of users in different emergency business scenarios, thereby providing users with accurate services. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0035] Figure 1 A flowchart of a method for extracting emergency domain intention information based on dependency grammar analysis provided by the present invention;

[0036] Figure 2 A schematic diagram of an electronic device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] like Figure 1 As shown, an embodiment of the present invention discloses a method for extracting emergency domain intention information based on dependency grammar analysis, comprising the following steps:

[0039] S1: Obtain the user's semantically complete text;

[0040] In one embodiment, S1 specifically includes:

[0041] Get the user's current input text;

[0042] The user's current input text, the user's historical input text, and the answers to the historical input text are input into the contextual multi-round understanding module to obtain the user's semantically complete text.

[0043] It can be understood that: the specific processing example of the context multi-round understanding module is given below:

[0044] Your task is to form a semantically complete question based on historical questions and answers and current input.

[0045] ##Require

[0046] 1. If you are currently inputting a question containing: this, these, these, the above, this, etc., do not cite specific accidents when completing the question. Just refer to the last question and this question to complete the question.

[0047] 2. If the current input is semantically complete, it will no longer be combined with the [historical question] and will directly return to the current [input] as the refined question

[0048] 3. If you cannot always refine the problem, return directly to [Input] as the refined problem

[0049] 4. Return the refined question directly without any explanation, and output it as a complete and clear question, with only one question mark in the whole sentence

[0050] 5. If there are many accidents listed in the previous answer, do not cite specific accidents when completing the question. Just refer to the previous question and this question to complete the question.

[0051] ##Example

[0052] ###Example 1

[0053] Historical issues:

[0054] Q: What is direct fire extinguishing?

[0055] A: Direct fire extinguishing is a method of extinguishing fires, which mainly uses forest fire trucks or other equipment to load chemical agents and spray them directly on the fire line to extinguish the fire.

[0056] enter:

[0057] Q: Indirect fire extinguishing

[0058] Output: What is indirect fire suppression?

[0059] ###Example 2

[0060] Historical issues:

[0061] Q: What are the dangerous chemicals that caused the current accident?

[0062] A: n-butane

[0063] enter:

[0064] Q: What fire extinguisher do you use?

[0065] Output:

[0066] What fire extinguisher is used for n-butane?

[0067] Example 3

[0068] Historical issues:

[0069] Q: How many major accidents occurred last year?

[0070] A: There were 5 major accidents last year.

[0071] enter:

[0072] Q: What are these accidents?

[0073] Output:

[0074] What major accidents occurred last year?

[0075] Example 4

[0076] Historical issues:

[0077] Q: How many disaster accidents occurred across the country in 2023?

[0078] A: According to the statistics of disaster accidents reported by the command center, there was 1 disaster accident in the country in 2023.

[0079] enter:

[0080] Q: How many people died in this accident?

[0081] Output:

[0082] How many people died from disasters in 2023?

[0083] S2: using a text classification model to classify the semantically complete text to obtain the intent of the semantically complete text;

[0084] In one embodiment, S2 specifically includes:

[0085] S31: preprocessing the semantically complete text;

[0086] S32: converting the preprocessed text into numerical features;

[0087] S33: Input the numerical features into a text classification model for classification to obtain a predefined intent corresponding to the semantically complete text; wherein the predefined intent corresponding to the semantically complete text is the intent of the semantically complete text.

[0088] In one embodiment, the preprocessing includes word segmentation and stop word removal;

[0089] Convert the preprocessed text into numerical features using bag-of-words (BOW), TF-IDF, or word embedding;

[0090] The text classification model is Naive Bayes or Support Vector Machine.

[0091] It is understandable that: taking the video dispatch scenario in emergency command and dispatch as an example (the corresponding intent types are different in different scenarios), the specific examples showing intent classification are as follows:

[0092] #Based on the content of multiple rounds of user input, determine which of the following 7 types of intentions the user's last round of input is intended for?

[0093] ##The intent categories are:

[0094] 1. Intent name: Open system; Intent description: Start or activate the specified software system or application. This operation does not involve specific page or element operations;

[0095] 2. Intent name: Exit / Close system; Intent description: Exit or close a specific system (electronic map system);

[0096] 3. Intent name: Switch monitoring screens; Intent description: Switch screens. The switch screen function allows users to adjust the display mode of display devices (such as computer monitors, TV screens, projectors, etc.) according to their needs, switching them from single-screen display to multi-screen combination display, or switching between different numbers of multi-screen combination displays.

[0097] 4. Intent name: Switch to single screen; Intent description: Switch to single screen;

[0098] 5. Intent name: Enlarge a certain monitoring; Intent description: Enlarge the monitoring;

[0099] 6. Intent name: Video description; Intent description: Provide a detailed description, summary and analysis of the surveillance camera, video or video stream with the specified serial number, including but not limited to its content, status, possible security risks, abnormal conditions or hidden dangers;

[0100] 7. Intent name: Open surveillance; Intent description: Open and view the surveillance camera or video at the specified location;

[0101] ##Please follow the following requirements:

[0102] 1. If the user's intention can be determined, output "Intent Name"; if the user's intention cannot be identified, only "Unidentified Intent" can be output.

[0103] 2. The output intent recognition result can only be "intent name" or "unrecognized intent".

[0104] ##Intent recognition example:

[0105] ###Example 1: The user's input content for multiple rounds is:

[0106] Round 1: Open the monitoring of the coal bunker of Changping Coal Industry in Golmud Oil Depot, Sizhuang Town, Gaoping City, Jincheng City, Shanxi Province

[0107] Round 2: Check out the main ventilation of Anjia

[0108] Round 3: A look at Busan Main Ventilation

[0109] Round 4: Check out the shift change location

[0110] Combining the context of multiple rounds of conversations, we can determine the user's intention of the "last round of input" as follows:

[0111] Open monitoring

[0112] According to the intention of the semantically complete text, a slot to be filled corresponding to the intention is obtained;

[0113] Performing dependency grammar analysis on the semantically complete text to obtain a dependency grammar analysis result of the semantically complete text;

[0114] It can be understood that dependency grammar analysis is a syntactic analysis method that focuses on the dependency relationship between words in a sentence rather than the parallel relationship between words. In dependency grammar analysis, a sentence is regarded as a collection of lexical units, each of which consists of a head and its dependents. The head is the core of the dominant meaning and syntactic function, while the dependents provide more information about the head.

[0115] In a certain embodiment, the present invention uses dependency analysis tools (such as Stanford Parser, SpaCy, etc.) to perform dependency grammar analysis on the emergency dispatch instruction Query to obtain the dependency relationship between each word and other words. For example, the core information in the dispatch instruction is extracted, including actions (such as dispatch, start, shut down, etc.), objects (such as systems, equipment, personnel, etc.) and related attributes or conditions (such as time, place, reason, etc.). And a tree structure is used to represent the structure of the entire sentence (such as "subject, predicate, object", "attributive, adverbial, complement", etc.). For example: the emergency dispatch instruction Query is "turn on the monitoring of Longxing Coal Mine in Hunan Province", and its dependency grammar analysis result is:

[0116]

[0117]

[0118]

[0119] The dependency grammar analysis results of the semantically complete text are assembled by engineering means to obtain the dependency grammar paraphrase results of the semantically complete text;

[0120] In a certain embodiment, based on the above dependency grammar analysis results, the above dependency grammar analysis results can be assembled by engineering means, and after removing non-core words, the above dependency grammar analysis results can be paraphrased as [opening" is a verb and is the core word of the instruction syntax, "Hunan Province" is a location word and is the attributive of "Longxing Coal Mine" in the instruction, "Longxing Coal Mine" is a location word and is the attributive of "monitoring" in the instruction, "monitoring" is a verb and is the predicate of "opening" in the instruction;]

[0121] S3: using the dependency grammar paraphrase result of the semantically complete text to perform similarity search in the example database to obtain an approximate example of the semantically complete text;

[0122] In one embodiment, similarity retrieval is performed based on the number and type of dependency relations in the dependency grammar paraphrase result of the semantically complete text.

[0123] S4: Acquire a prompt word template; wherein the prompt word template includes an example step construction; the example step construction includes the approximate example and the dependency grammar paraphrase result of the approximate example;

[0124] In a certain embodiment, the prompt word template also includes a model role definition, an intent information extraction slot definition, an approximate example intent information extraction result, and a task requirement.

[0125] In one embodiment, the task requires that the final output intent information extraction result be in the format of json.

[0126] S5: Input the slot to be filled, the prompt word template and the semantically complete text into the large language model to obtain the intention information extraction result of the user's current input text.

[0127] It can be understood that: the large language model extracts slot information from the semantically complete text according to the prompt word template, and fills the extracted slot information into the slot to be filled, thereby obtaining the intention information extraction result of the user's current input text;

[0128] The following is an example of using a large language model to extract slot information (i.e., entity extraction):

[0129] #Combining the information input by the user in multiple rounds, extract the content corresponding to the entity in the last round of user input.

[0130] ##The entity objects that need to be extracted are:

[0131] 1. Entity name: province; Entity description: China's first-level administrative region, provincial level, including provinces, autonomous regions, municipalities directly under the central government, and special administrative regions; variable name: province; for example: Heilongjiang Province, Jilin Province, Liaoning Province, Beijing, Guangdong Province, Inner Mongolia Autonomous Region.

[0132] 2. Entity name: city; Entity description: China's second-level administrative region, city level, including prefecture-level cities, regions, leagues, autonomous prefectures, etc.; Variable name: city; for example: Guangzhou, Foshan, Urumqi, Chengdu, Hangzhou.

[0133] 3. Entity name: District / County; Entity description: China's third-level administrative region, district and county level, including counties, autonomous counties, urban districts, county-level cities, etc.; Variable name: district; for example: Chaoyang District, Wuqing District, Qingyang District, Haishu District, Yuelu District.

[0134] 4. Entity name: town / village / street; Entity description: China's fourth-level administrative region, township level, including townships, towns, and streets; variable name: town; for example: Chaoyang Street, Xinhua Town, Baiyun Township, Xinglong Street, Nanshan Town.

[0135] 5. Entity name: village / community; Entity description: Village and community are basic administrative units, including a certain number of family residences and community facilities; Variable name: village; For example: Shanxin Village, Jianshe Street Community.

[0136] 6. Entity name: place; Entity description: the specific location name or location type where the event occurred, including: company name, school name, landmark name, etc. or school, hospital, mine, etc.; variable name: place; for example: Wuni Ceramic Clay Mine, First People's Hospital.

[0137] 7. Entity name: point; Entity description: Specific location subordinate to the location, including: door, classroom, workshop and other specific locations; Variable name: position; For example: workshop-300-14, loading platform 4-10 on the north side, tank area, main entrance, parking lot.

[0138] 8. Entity name: event name; Entity description: Real-time dynamic information of the monitoring site, including personnel working, on-duty, on-call, operating, and operating conditions; vehicle entry, exit, arrival, departure, visit, visit, and passing dynamics; as well as emergency conditions such as fire occurrence, fire spread, and fire incidents. Comprehensively identify the real-time status and changes of personnel, vehicles, and fires at the scene. ; Variable name: event_name; For example: someone is working, someone is on-call, a vehicle is visiting, on fire, explosion, collapse.

[0139] 9. Entity name: whether the event subject exists; Entity description: Determine whether the subject in "event name" exists, value type: boolean, if it exists, return: True, if it does not exist, return: False. ; Variable name: event_subject_exist; For example: 'someone working' returns 'True', 'vehicles entering and exiting' returns 'True', 'fire' returns 'True', 'staff not at work' returns 'False', 'staff leaving' returns 'False', 'no vehicle detected' returns 'False', 'fire extinguished' returns

[0140] 'False'.

[0141] ##Requirements that entity extraction must follow:

[0142] 1. You must think according to the steps of the example and draw conclusions step by step.

[0143] 2. The results of entity extraction are always output in JSON format.

[0144] 3. If the entity cannot be extracted, the value of the corresponding entity variable is an empty string, that is: "".

[0145] 4. The output JSON format of the entity extraction result is: {"province":"****","city":"****","district":"****","town":"****","village":"****","place":"****","position":"****","event_name":"****",

[0146] "event_subject_exist":"****"}.

[0147] ##Entity extraction example:

[0148] ###Example 1: Retrieve the monitoring of Shimentou Nature Reserve in Qingxin District, Qingyuan;

[0149] The analysis steps are as follows:

[0150] 1. Word segmentation and syntactic analysis: "Call" is a verb and the core word of the command syntax; "Qingyuan" is a location word and the attributive of "Qingxin District" in the command; "Qingxin District" is a location word and the attributive of "Shimentou Protection Area" in the command; "Shimentou Protection Area" is a location word and the attributive of "monitoring" in the command; "monitoring" is a verb and the predicate of "call" in the command; 2. Extract core parameters: Intention: call monitoring, administrative regions: Qingyuan, Qingxin District; location: Shimentou Protection Area, monitoring point: none;

[0151] The final output is:

[0152] {"province":"","city":"Qingyuan","district":"Qingxin District","town":"","village":"","place":"Mentou Protection Area","position":"","event_name":"","event_subject_exist":""}

[0153] ##The user's input content for multiple rounds is:

[0154] Question: Open the surveillance of Longxing Coal Mine in Hunan Province.

[0155] 1. Word segmentation and syntactic analysis: "open" is a verb and is the core word of the instruction syntax; "Hunan Province" is a location word and is the attributive of "Longxing Coal Mine" in the instruction; "Longxing Coal Mine" is a location word and is the attributive of "monitor" in the instruction; "monitor" is a verb and is the predicate of "open" in the instruction;

[0156] In one embodiment, the prompt word template is specifically shown in the following table:

[0157]

[0158]

[0159]

[0160] The present invention not only utilizes the powerful processing power of large language models, but also guides the model to generate more accurate and explainable answers through Chain-of-Thought (CoT) technology and step-by-step reasoning process. This method breaks down complex problems into a series of simple sub-problems by clarifying the thinking process of each step, and injects sample instructions similar to the query and its dependency grammar analysis conclusions into the prompt words, so that the large model can more accurately understand and follow specific instructions in the emergency field.

[0161] Finally, it should be noted that:

[0162] The present invention evaluates the intention information extraction method provided by the present invention from three dimensions: the fitting degree of the thinking chain reasoning steps, the accuracy of the output result format, and the accuracy of the intention information extraction result. The fitting degree of the reasoning steps and the accuracy of the output result format are mainly used for the large model instruction following effect, whether there is no longer permission for past instructions after multiple reasonings; and the accuracy of the output result content is used to evaluate the final result;

[0163] The fitting degree of the reasoning steps of the thinking chain: The derivation process of the thinking chain includes three steps: query segmentation, syntactic analysis, and information extraction. The current detection method only evaluates the results of large-scale model reasoning, but not the accuracy of the content.

[0164] Output result format accuracy: To facilitate the smooth execution of subsequent scheduling projects, we require the final intent information extraction results to be output in JSON format in the prompt word project. The evaluation criterion is whether the final inference result of the large model outputs the extraction results in JSON format.

[0165] Accuracy of intent information extraction results: The present invention mainly uses identification personnel to first construct a test data set, which contains test Query instructions and corresponding intent slot contents, and manually compares and scores the output results of each large model, and calculates the accuracy of intent information extraction according to the following two dimensions.

[0166] Slot extraction error rate

[0167] The slot extraction error rate (SER) calculation formula is as follows:

[0168] SER = Number of slots with extraction errors / Total number of slots

[0169] Command sentence error rate

[0170] The formula for calculating the whole sentence error rate (QER) is as follows:

[0171] QER = Number of instructions with slot fetch errors / Total number of instructions

[0172] The present invention further ensures the accuracy and reliability of instruction compliance by evaluating the fitting degree of the reasoning steps of the thinking chain, the accuracy of the output result format, and the accuracy of the intention information extraction result. This method not only improves the accuracy of intention recognition, but also ensures that the final output result can meet the needs of actual operations in the emergency field, thus having obvious advantages in instruction compliance.

[0173] In summary, compared with the traditional BERT-based natural language processing (NLP) algorithm, the present invention significantly improves the generalization of intent recognition by combining dependency grammar analysis and large model COT (Chain-Of-Thought) technology. Although traditional NLP algorithms have achieved remarkable results in many fields, in the emergency field, due to the widespread use of professional terms and special contexts, it is often difficult to accurately cover and understand these special situations. The method of the present invention can more accurately parse the sentence structure and the dependency relationship between words through dependency grammar analysis, so as to better understand the professional language environment in the emergency field. In addition, the large model COT technology enhances the migration ability of the model by training on similar data sets, so that the algorithm can adapt to different emergency business scenarios, such as fire, earthquake, etc., which has obvious advantages in generalization.

[0174] Compared with the general method of directly identifying intent through a large model, the present invention performs better in terms of instruction following. The present invention not only utilizes the powerful processing power of a large language model, but also guides the model to generate more accurate and explainable answers through the Chain-of-Thought (CoT) technology and the step-by-step reasoning process. This method breaks down complex problems into a series of simple sub-problems by clarifying the thinking process of each step, and injects example instructions similar to the Query and its dependency grammar analysis conclusions into the prompt words, so that the large model can more accurately understand and follow specific instructions in the emergency field. In addition, by evaluating the fitting degree of the reasoning steps of the chain of thought, the accuracy of the output result format, and the accuracy of the intent information extraction results, the accuracy and reliability of instruction following are further ensured. This method not only improves the accuracy of intent recognition, but also ensures that the final output result can meet the needs of actual operations in the emergency field, thus having obvious advantages in instruction following.

[0175] Figure 2 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 2As shown, the electronic device may include: a processor 201, a communication interface 202, a memory 203 and a communication bus 204, wherein the processor 201, the communication interface 202 and the memory 203 communicate with each other through the communication bus 204. The processor 201 may call the logic instructions in the memory 203 to execute the emergency domain intention information extraction method based on dependency grammar analysis.

[0176] In addition, the logic instructions in the above-mentioned memory 203 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0177] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the emergency field intention information extraction method based on dependency grammar analysis provided by the above-mentioned methods.

[0178] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the emergency domain intention information extraction method based on dependency grammar analysis provided by the above-mentioned methods.

[0179] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0180] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0181] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0182] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for extracting emergency domain intention information based on dependency grammar analysis, characterized in that: The following steps are involved: S1: Get the user's semantically complete text; S2: using a text classification model to classify the semantically complete text to obtain the intent of the semantically complete text; According to the intention of the semantically complete text, a slot to be filled corresponding to the intention is obtained; Performing dependency grammar analysis on the semantically complete text to obtain a dependency grammar analysis result of the semantically complete text; The dependency grammar analysis results of the semantically complete text are assembled by engineering means to obtain the dependency grammar paraphrase results of the semantically complete text; S3: using the dependency grammar paraphrase result of the semantically complete text to perform similarity search in the example database to obtain an approximate example of the semantically complete text; S4: Acquire a prompt word template; wherein the prompt word template includes an example step construction; the example step construction includes the approximate example and the dependency grammar paraphrase result of the approximate example; S5: Input the slot to be filled, the prompt word template and the semantically complete text into the large language model to obtain the intention information extraction result of the user's current input text.

2. According to the method for extracting emergency domain intention information based on dependency grammar analysis according to claim 1, it is characterized in that: S1 specifically includes: Get the user's current input text; The user's current input text and the user's historical input text are input into the contextual multi-round understanding module to obtain the user's semantically complete text.

3. According to the method for extracting emergency domain intention information based on dependency grammar analysis according to claim 1, it is characterized in that: S2 specifically includes: S31: preprocessing the semantically complete text; S32: converting the preprocessed text into numerical features; S33: Input the numerical features into a text classification model for classification to obtain a predefined intent corresponding to the semantically complete text; wherein the predefined intent corresponding to the semantically complete text is the intent of the semantically complete text.

4. According to the method for extracting emergency domain intention information based on dependency grammar analysis according to claim 3, it is characterized in that: The preprocessing includes word segmentation and stop word removal; Convert the preprocessed text into numerical features using bag-of-words (BOW), TF-IDF, or word embedding; The text classification model is Naive Bayes or Support Vector Machine.

5. According to the method for extracting emergency domain intention information based on dependency grammar analysis according to claim 1, it is characterized in that: Similarity retrieval is performed based on the number and type of dependency relations in the dependency grammar paraphrase result of the semantically complete text.

6. According to the method for extracting emergency domain intention information based on dependency grammar analysis according to claim 1, it is characterized in that: The prompt word template also includes model role definition, intent information extraction slot definition, intent information extraction results of approximate examples and task requirements.

7. The method for extracting emergency domain intention information based on dependency grammar analysis according to claim 6 is characterized in that: The task requires that the final output intent information extraction result be in the format of JSON.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the emergency field intention information extraction method based on dependency grammar analysis as described in any one of claims 1 to 7.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the emergency domain intention information extraction method based on dependency grammar analysis as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the emergency domain intention information extraction method based on dependency grammar analysis as described in any one of claims 1 to 7.

Images