Method, device and equipment for converting vertical class demand based on multi-modal term normalization

Abstract

The application provides a multi-modal term normalization-based vertical class demand conversion method, device and equipment, and relates to the technical field of data processing. The method comprises the following steps: extracting a plurality of demand terms from a to-be-converted text, and determining a demand term word vector corresponding to each demand term; the to-be-converted text is obtained by text processing of multi-modal vertical class demands; determining a vertical class of each to-be-converted text according to the demand term word vectors of the plurality of demand terms of each to-be-converted text; inputting the demand term word vector of each to-be-converted text into a term normalization model corresponding to the vertical class of the to-be-converted text, to obtain a technical term corresponding to each demand term; and determining a demand conversion result of the multi-modal vertical class demand based on the technical term corresponding to each demand term. The application can effectively solve the problem of industry term difference, and improve the text conversion efficiency and accuracy.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, and equipment for transforming vertical requirements based on multimodal terminology normalization. Background Technology

[0002] In various vertical sectors such as industry, healthcare, and finance, the healthcare sector includes treatment scenarios for oncology, Alzheimer's disease, and cancer prevention, while the industrial sector includes blasting scenarios. Business experts focus on the business itself, accumulating a large amount of industry-specific terminology to accurately describe business phenomena, processes, and requirements, while technical personnel focus on technical implementation, using technical terminology to build and optimize solutions. Therefore, a significant terminology gap exists between business experts and technical personnel. Furthermore, their different knowledge systems and experience backgrounds lead to differences in their understanding and expression of the same concepts, further widening the terminology gap and making it difficult for them to accurately understand each other's intentions during communication and collaboration.

[0003] In existing technologies, building a terminology database is an effective way to bridge the terminology gap between business experts and technical personnel in vertical scenarios. Through in-depth communication with business experts, commonly used requirement terms in specific fields and their meanings in different scenarios are collected and organized. Technical personnel, combining their own knowledge and project experience, provide technical interpretations of these requirement terms, clarifying their corresponding technical concepts. These requirement terms and their definitions are then compiled into a structured terminology database for reference during communication between both parties. When inconsistencies arise in terminology, both parties can consult the database to ensure consistent understanding, thereby reducing misunderstandings and communication barriers, and improving communication efficiency and the smoothness of project collaboration.

[0004] However, terminology databases may suffer from problems such as outdated terminology and unclear terminology meanings in different scenarios during use. Furthermore, the workload of manually consulting and comparing terminology databases is relatively large, resulting in low terminology conversion efficiency. Summary of the Invention

[0005] This invention provides a method, apparatus, and device for vertical demand conversion based on multimodal terminology normalization, in order to solve the problems that terminology databases may encounter during use, such as untimely terminology updates, unclear terminology meanings in different scenarios, and low terminology conversion efficiency.

[0006] In a first aspect, embodiments of the present invention provide a method for transforming vertical requirements based on multimodal terminology normalization, including:

[0007] Multiple requirement terms are extracted from the text to be transformed, and the corresponding requirement term word vector is determined for each requirement term; the text to be transformed is obtained by textualizing multimodal vertical requirements.

[0008] Based on the demand term word vectors of multiple demand terms for each text to be converted, determine the vertical category of each text to be converted;

[0009] Input the term vector of the demand term for each text to be converted into the term normalization model corresponding to the vertical category of the text to be converted to obtain the corresponding technical term for each demand term.

[0010] Based on the corresponding technical terms for each requirement term, the requirement transformation results for multimodal vertical requirements are determined.

[0011] Secondly, embodiments of the present invention provide a vertical demand transformation device based on multimodal terminology normalization, comprising:

[0012] The extraction module is used to extract multiple requirement terms from the text to be transformed and determine the corresponding requirement term word vector for each requirement term; the text to be transformed is obtained by textualizing multimodal vertical requirements;

[0013] The category determination module is used to determine the vertical category of each text to be converted based on the demand term word vectors of multiple demand terms for each text to be converted;

[0014] The conversion module is used to input the word vector of the demand term in each text to be converted into the terminology normalization model corresponding to the vertical category of the text to be converted, so as to obtain the corresponding technical terminology for each demand term.

[0015] The result determination module is used to determine the requirement transformation result of multimodal vertical requirements based on the corresponding technical terms of each requirement term.

[0016] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.

[0017] In this embodiment of the invention, multiple requirement terms are extracted from the text to be converted obtained by textualizing the multimodal vertical requirements to be converted. The term vectors for each requirement term are further determined, and the vertical category is further determined. A terminology normalization model is determined based on the vertical category, enabling accurate conversion of requirement terms according to different vertical categories, ensuring the accuracy of the conversion results. The term vectors of the requirement terms are input into the terminology normalization model of the corresponding vertical category of the text to be converted to obtain the corresponding technical terms for each requirement term. Based on the technical terms of each requirement term, the requirement conversion result for the multimodal vertical requirement is determined. This effectively solves the problem of industry terminology differences and improves the efficiency and accuracy of text conversion. Attached Figure Description

[0018] Figure 1This is a flowchart illustrating the implementation of the vertical requirement transformation method based on multimodal terminology normalization provided in this embodiment of the invention.

[0019] Figure 2 This is a flowchart illustrating the implementation of step S130 of the vertical demand transformation method based on multimodal terminology normalization provided in this embodiment of the invention.

[0020] Figure 3 This is a flowchart illustrating the implementation of step S1304 of the vertical demand transformation method based on multimodal terminology normalization provided in this embodiment of the invention.

[0021] Figure 4 This is a schematic diagram of the structure of the vertical demand conversion device based on multimodal terminology normalization provided in an embodiment of the present invention;

[0022] Figure 5 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0023] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] See Figure 1 The document illustrates a flowchart of the implementation of the vertical demand transformation method based on multimodal terminology normalization provided in this embodiment of the invention, detailed below:

[0025] Step S110: Extract multiple requirement terms from the text to be converted and determine the corresponding requirement term word vector for each requirement term; the text to be converted is obtained by textualizing multimodal vertical requirements.

[0026] In some embodiments, the text to be converted refers to the text content obtained after textualizing multimodal vertical requirements. For example, if a multimodal vertical requirement is a piece of equipment operation voice in an industrial field, the text describing the operation process obtained after processing by a speech recognition model is the text to be converted. Requirement terms are industry-specific terms used by business experts in specific vertical scenarios to accurately describe business phenomena, processes, and requirements. Requirement term word vectors are numerical vectors that transform requirement terms into a series of terms that reflect their semantic information, facilitating computer understanding and processing of the meaning of the term. In multimodal vertical requirements, multimodality refers to information in multiple forms, such as voice, text, and video; vertical refers to a specific industry or field, such as finance or industry. For example, a customer consultation voice, a financial product description text, and a stock market analysis video in the financial field together constitute a multimodal vertical requirement in the financial field. Textualization is the process of converting non-textual multimodal requirements into text.

[0027] In one possible implementation, the specific method for text-based processing of multimodal vertical requirements is as follows: inputting the speech requirements in the multimodal vertical requirements into a preset speech recognition model to obtain the corresponding text to be converted; determining the text requirements in the multimodal vertical requirements as the corresponding text to be converted; extracting audio from the video requirements in the multimodal vertical requirements to obtain the corresponding audio, and inputting the audio into a preset speech recognition model to obtain the corresponding text to be converted.

[0028] In some embodiments, voice requirements are requirements that exist in voice form within a multimodal vertical requirement. For example, in an industrial setting, a technician's verbal description of equipment malfunctions constitutes a voice requirement. A pre-set speech recognition model refers to a pre-defined model used to convert speech information into text information. A model that can convert a spoken phrase like "equipment temperature is too high" into corresponding text can be used as a pre-set speech recognition model. Text requirements are requirements that exist in text form within a multimodal vertical requirement. In the medical field, the handwritten text content on a doctor's diagnosis record of a patient's condition is a text requirement. Video requirements are a type of multimodal vertical requirement that are presented in video form. For example, in the financial field, a video recording a customer explains their investment questions, which is a video requirement. Audio extraction refers to the process of separating sound information from video. Extracting the speaker's voice from a video containing product introductions is audio extraction. Audio refers to the sound information extracted from video and can be used as input to a speech recognition model. For example, the teacher's voice extracted from an instructional video is audio.

[0029] In one possible implementation, before step S110, the method further includes: for each vertical category, training a deep learning model based on the demand term word vectors of demand terms, the annotation results of each demand term, and the technical term word vectors of each technical term in each term sample of the corresponding vertical category, to obtain a term normalization model corresponding to the vertical category; wherein, the term samples are manually annotated term samples; and the annotation results of each demand term are the corresponding technical terms for each demand term.

[0030] In some embodiments, a terminology sample refers to a manually annotated sample containing demand terms and their corresponding information. In the financial field, a manually compiled and annotated sample of "annualized return" (a demand term) and its related information constitutes a terminology sample. Annotation results refer to the corresponding technical terms manually annotated for each demand term. For example, manually annotating the financial demand term "annualized return" with its corresponding technical term "annual rate of return" is the annotation result for that demand term. Technical terms are terms used by technical personnel when building and optimizing solutions to accurately describe technical concepts and implementation methods. For example, in the industrial field, the term "rated power" used by technical personnel to describe equipment operating parameters is a technical term. Technical term word vectors are numerical vectors that transform technical terms into a series of values ​​that reflect their semantic information, facilitating computer processing and understanding of the meaning of the technical term. For example, the technical term "rated power" is transformed into a set of numerical vectors reflecting its association with concepts such as "power limit" and "equipment energy consumption." A deep learning model is a machine learning model based on multi-layer neural networks that can achieve specific functions by learning patterns in large amounts of data. For example, a neural network model that can gradually master the correspondence between demand terms and technical terms by learning from a large number of terminology samples is a deep learning model in this context. A terminology normalization model is a model trained with deep learning for a specific vertical category. It is used to convert demand terms into corresponding technical terms. For example, a model trained for the industrial field that can convert "machine output" (demand term) into "equipment output power" (technical term) is a terminology normalization model for the industrial vertical.

[0031] Step S120: Determine the vertical category of each text to be converted based on the demand term word vectors of multiple demand terms for each text to be converted.

[0032] In some embodiments, a vertical category refers to a specific industry or field category, such as different professional fields like industry, healthcare, and finance. If the demand terms in the text to be converted are mostly medical-related words such as "myocardial infarction" and "prescription drugs," then the vertical category corresponding to the text to be converted may be the healthcare field; if the demand terms are mostly "stock trends" and "financial products," then the corresponding vertical category may be the financial field.

[0033] In one possible implementation, step S120 is specifically processed as follows: For each text to be converted, the following steps are performed: Calculate the distance between each term vector in the term vector library of the first vertical category and the first demand term word vector of the first demand term in the text to be converted, and determine the minimum value among the distances as the first distance value between the first demand term and the first vertical category; wherein, the first vertical category is any vertical category; the first demand term is any demand term; add the first distance values ​​of the multiple first demand terms of the text to be converted to the first vertical category to obtain the second distance value between the text to be converted and the first vertical category; determine the vertical category with the smallest second distance value with the text to be converted as the vertical category of the text to be converted.

[0034] In some embodiments, the first vertical category refers to any single vertical category, that is, a specific field category randomly selected from all vertical categories. Among the vertical categories of industry, healthcare, and finance, healthcare is chosen as the analysis object; in this case, healthcare is a first vertical category. The terminology vector library refers to the collection of word vectors corresponding to the technical terms in each vertical category, used to provide a reference for calculating the distance between demand terms and vertical categories. For example, the terminology vector library for the healthcare vertical category may contain word vectors for technical terms such as "myocardial infarction" and "prescription drugs," while the terminology vector library for the finance vertical category may contain word vectors for technical terms such as "annual rate of return" and "asset allocation." A terminology vector refers to the word vector corresponding to the technical term stored in the terminology vector library; it is a numerical vector that reflects the semantic information of the technical term. For example, the word vector corresponding to "annual rate of return" in the finance vertical category terminology vector library is a terminology vector. The first requirement term refers to any one requirement term in the text to be transformed. That is, it's any one selected from multiple requirement terms extracted from the text. If the text contains requirement terms such as "machine output" and "parts wear," then "machine output" is a first requirement term when it's selected for analysis. The first requirement term word vector refers to the word vector corresponding to the first requirement term, that is, the numerical vector that reflects the semantic information of the first requirement term after transformation. For example, when the first requirement term is "machine output," the corresponding numerical vector reflecting the semantics related to "equipment output capacity" is the first requirement term word vector.

[0035] In some embodiments, distance refers to the similarity metric between the word vector of the first demand term and the term vectors in the term vector library. The smaller the distance, the closer the terms represented by the two vectors are semantically. The word vector of "machine output" has a smaller distance from the word vector of "equipment output power" in the industrial vertical term vector library, while it has a larger distance from the word vector of "heart rate" in the medical vertical term vector library. The first distance value refers to the minimum distance between the first demand term and the first vertical category, that is, the minimum distance among all term vectors in the term vector library of the first demand term and the first vertical category. When the industrial vertical category is the first vertical category, after calculating the distance between the word vector of "machine output" and all term vectors in the industrial vertical term vector library, the smallest distance value is the first distance value between "machine output" and the industrial vertical category. The second distance value refers to the total distance between the text to be converted and the first vertical category. It is obtained by adding the first distance values ​​of all the first demand terms in the text to be converted to the first vertical category. When the industrial vertical category is the first vertical category, a text to be converted contains 3 demand terms, and the first distance values ​​of each term with the industrial vertical category are 0.2, 0.3, and 0.1, respectively. Then the second distance value between the text and the industrial vertical category is 0.6.

[0036] Step S130: Input the word vector of the demand term for each text to be converted into the terminology normalization model corresponding to the vertical category of the text to be converted, and obtain the technical terminology corresponding to each demand term.

[0037] In some embodiments, input refers to the operation of passing the word vectors of demand terms to the terminology normalization model, which is a prerequisite for the model to process the data. For example, passing the word vector of the demand term "machine output" in the text to be transformed in the industrial field to the terminology normalization model corresponding to the industrial vertical is the input process. Obtaining the corresponding technical term for each demand term refers to the process of obtaining the corresponding result after processing the input demand term word vectors through the terminology normalization model. When the word vector of "machine output" is input into the industrial terminology normalization model, the model calculates and derives the corresponding technical term.

[0038] See Figure 2 The specific processing method of step S130 above includes steps S1301-S1304, the details of which are as follows:

[0039] Step S1301: Input the word vector of the demand term for each text to be converted into the term normalization model corresponding to the vertical category of the text to be converted, and obtain at least one initial technical term for each demand term.

[0040] In some embodiments, "at least one initial technical term" means that there are at least one initial technical term. This could be one or more. For example, after processing by the model, a demand term might correspond to one initial technical term, or two or more. Initial technical terms refer to the technical terms output by the terminology normalization model after preliminary processing of the demand term's word vector. They are not the final technical terms and require further filtering. For instance, when the financial demand term "annualized return" is input into the model, the model might initially output "annual rate of return" and "annualized rate of return," which are the initial technical terms for that demand term.

[0041] Step S1302: Determine the number of occurrences of each requirement term in the text to be converted, and sort each requirement term in descending order of the number of occurrences.

[0042] In some embodiments, the frequency of occurrence refers to the number of times a requirement term is mentioned or appears in the text to be converted. In the text to be converted in the industrial field, the requirement term "machine vibration" is mentioned multiple times, and this specific number of mentions is its frequency of occurrence. Ordering by frequency of occurrence from highest to lowest: This refers to arranging the requirement terms from the most frequent to the least frequent based on their frequency of occurrence. For example, if "data anomaly" appears 6 times, "system lag" appears 4 times, and "network interruption" appears 2 times in the text to be converted, then the order from highest to lowest frequency would be "data anomaly," "system lag," and "network interruption."

[0043] Step S1303: Determine the weight of each requirement term based on the ranking result of each requirement term.

[0044] In some embodiments, the ranking result refers to the order in which requirement terms are arranged according to their frequency of occurrence from largest to smallest. For example, if "data anomaly" appears 6 times, "system lag" appears 4 times, and "network interruption" appears 2 times in the text to be converted, then the order from largest to smallest is "data anomaly," "system lag," and "network interruption." This sequence list is the ranking result of these requirement terms. Determining the weight of each requirement term refers to the process of clarifying the importance of each requirement term based on the ranking result. For example, based on the ranking result of "data anomaly," "system lag," and "network interruption," the importance value of each term in the overall requirement is determined. The requirement term that appears earlier in the ranking has a larger weight. For example, if "data anomaly" appears 6 times, "system lag" appears 4 times, and "network interruption" appears 2 times in the text to be converted, then the weight of "data anomaly" is 6 ÷ (6 + 4 + 2) × 100% = 0.5.

[0045] Step S1304: Based on the weight of each demand term and at least one initial technical term corresponding to each demand term, determine the corresponding technical term for each demand term.

[0046] In some embodiments, determining the corresponding technical term for each demand term refers to the process of combining the weight of each demand term with at least one corresponding initial technical term, and then analyzing and filtering to determine the final corresponding technical term. For example, in the industrial field, combining the weight of "machine vibration" with its initial technical terms "equipment oscillation" and "mechanical vibration," the best match is determined as the final technical term through analysis. The corresponding technical term for each demand term refers to the technical term that precisely corresponds to the demand term after comprehensive analysis of the weights and initial technical terms. It is the final result after the demand term is transformed. For example, in the medical field, the demand term "heart failure" is combined with its weights and the initial technical terms "decreased cardiac function" and "cardiac dysfunction," and "decreased cardiac function" is finally determined as its corresponding technical term. This "decreased cardiac function" is the technical term corresponding to the demand term.

[0047] See Figure 3 The specific processing method of step S1304 above includes steps S13041-S1302, the details of which are as follows:

[0048] Step S13041: If there is a corresponding initial technical term for the first requirement term, then the initial technical term is determined as the technical term of the first requirement term; wherein, the first requirement term is any requirement term.

[0049] In some embodiments, when the number of initial technical terms corresponding to the first demand term is a single one, the initial technical term is determined as the technical term of the first demand term. For example, if the first demand term "lung infection" in the medical field is processed by the model and only the initial technical term "lung infection" is obtained, then "lung infection" is determined as the technical term of "lung infection".

[0050] Step S13042: If there are multiple corresponding initial technical terms for the first demand term, then perform the following steps: determine the first confidence level of each technical term based on the distance between the demand term word vector of the first demand term and the demand term word vector of each initial technical term; determine the second confidence level of each technical term based on the semantic matching degree of each initial technical term of the first demand term; determine the technical term of the first demand term based on the weight of each first demand term, the first confidence level and the second confidence level of each initial technical term corresponding to the first demand term.

[0051] In some embodiments, the first confidence level refers to an indicator calculated based on the distance between the word vector of the first demand term and the word vector of the demand term of each initial technical term. This indicator measures the degree of trustworthiness in the match between the initial technical term and the first demand term. The smaller the distance between the word vectors, the higher the first confidence level. For example, the word vector of "unstable machine operation" is closer to the word vector of "fluctuating equipment operation," so the first confidence level of "fluctuating equipment operation" is higher. Semantic matching degree refers to the degree of semantic fit between the first demand term and each initial technical term. The higher the fit, the higher the semantic matching degree. For example, in the medical field, the first demand term "difficulty breathing" is semantically closer to the initial technical term "breathing difficulties," resulting in a higher semantic matching degree; while "lung discomfort" has a lower fit and therefore a lower semantic matching degree. The second confidence level is an indicator used to measure the credibility of the match between the initial technical term and the first requirement term, determined based on the semantic matching degree between the first requirement term and each initial technical term. The higher the semantic matching degree, the higher the second confidence level. For example, if the semantic matching degree between "difficulty breathing" and "breathing difficulties" is high, then the second confidence level of "breathing difficulties" will be relatively high. The second confidence level of "breathing difficulties" can be obtained by dividing the semantic matching degree of "breathing difficulties" by the sum of the semantic matching degree of "breathing difficulties" and the semantic matching degree of "lung discomfort".

[0052] It should be noted that when there are two or more initial technical terms for the first requirement term, the first confidence level of each initial technical term needs to be determined based on the distance between each initial technical term and the requirement term word vector of the first requirement term. For example, if the first requirement term A has three corresponding initial technical terms A1, A2, and A3, and the distance between A and A1 is d1, the distance between A and A2 is d2, and the distance between A and A3 is d3, then the first confidence level of A1 is... .

[0053] In one possible implementation, step S13042 is specifically processed as follows: if the weight of the first requirement term exceeds the preset weight, then the initial technical term with the highest first confidence level is determined as the technical term of the first requirement term; if the weight of the first requirement term does not exceed the preset weight, then the sum of the first confidence level and the second confidence level of each technical term is determined as the third confidence level, and the initial technical term with the highest third confidence level is determined as the technical term of the first requirement term.

[0054] In some embodiments, a preset weight refers to a pre-defined weight threshold used to measure whether the importance of the first demand term reaches a specific standard. For example, in the financial field, a preset weight may be set for demand terms; when the weight of a demand term exceeds this value, it indicates that it is more critical in demand conversion. The initial technical term with the highest first confidence level refers to the initial technical term with the highest first confidence value among multiple initial technical terms. For example, the first demand term "slow cash flow" has two initial technical terms: "slow cash turnover" and "slow cash flow." "Slow cash turnover" has a higher first confidence level, so it is the initial technical term with the highest first confidence level. The weight of the first demand term not exceeding the preset weight means that the weight of the first demand term is less than or equal to the preset weight, indicating that the importance of this demand term in the overall demand is relatively low. If the weight of a demand term is 0.3, and the preset weight is 0.5, then the weight of this demand term does not exceed the preset weight. The third confidence level is the value obtained by adding the first confidence level and the second confidence level. It is an indicator that comprehensively measures the credibility of the match between the initial technical term and the first requirement term. If the first confidence level of an initial technical term is 0.4 and the second confidence level is 0.3, the sum of the two is 0.7, and 0.7 is the third confidence level of the initial technical term.

[0055] It should be noted that among multiple initial technical terms, the initial technical term with the highest third confidence score is determined as the technical term. For example, the first requirement term "equipment heating" has two initial technical terms: "equipment temperature too high" and "device heating abnormally". The former has a third confidence score of 0.8 and the latter has a score of 0.6. Therefore, "equipment temperature too high" is the technical term for the first requirement term "equipment heating".

[0056] By inputting the word vectors of demand terms into the terminology normalization model of the corresponding vertical category, it is possible to ensure that the conversion is aligned with the characteristics of specific fields and improve adaptability. By generating at least one initial technical term, the limitations of a single result are avoided, providing more possibilities for subsequent screening. Determining the weight based on the frequency of occurrence highlights the importance of high-frequency demand terms, allowing key information to be prioritized in the conversion process. Handling the number of initial technical terms in different ways can improve processing efficiency while ensuring the accuracy of key term conversion. In addition, by combining the first and second confidence levels to determine technical terms, and by conducting multi-dimensional evaluation of each initial technical term, the reasonableness of the conversion results can be improved while ensuring the accuracy and flexibility of the conversion. This effectively solves the potential one-sidedness problem in terminology conversion and improves the overall conversion quality.

[0057] Step S140: Based on the corresponding technical terms of each requirement term, determine the requirement transformation result of the multimodal vertical requirements.

[0058] In some embodiments, the requirement transformation result refers to the requirement content that can be accurately understood by technical personnel after the requirement terms in the multimodal vertical requirement are converted into corresponding technical terms. It is the final output of the entire transformation process. For example, in the medical field, a multimodal requirement may contain the voice description of "heart failure" and the text record of "difficulty breathing." After transformation, the requirement information may include technical terms such as "decreased cardiac function" and "dyspnea." This information is the requirement transformation result of that multimodal vertical requirement.

[0059] In one possible implementation, step S140 is specifically processed as follows: each requirement term in the text to be converted is replaced with the corresponding technical term to obtain the text conversion result of the text to be converted; multiple requirement information containing technical terms are extracted from the text conversion result, and all requirement information of the sample to be converted is determined as the requirement conversion result of the corresponding multimodal vertical requirement.

[0060] In some embodiments, substitution refers to the operation of replacing the original requirement terms in the text to be converted with their corresponding technical terms. For example, in the text to be converted in the industrial field, replacing "severe machine vibration" with "large equipment vibration amplitude" is a substitution process. The text conversion result refers to the new text formed after the text to be converted has undergone the substitution operation, in which all requirement terms have been replaced by corresponding technical terms. For example, the original text to be converted is "insufficient machine output, rapid parts wear," and after substitution it becomes "insufficient equipment output power, rapid component wear." This new text is the text conversion result of the text to be converted. Requirement information containing technical terms refers to the specific requirement content containing technical terms extracted from the text conversion result. For example, "decreased cardiac function requires drug intervention" and "dyspnea requires oxygen therapy support" extracted from the text conversion result in the medical field are multiple pieces of requirement information containing technical terms. After extracting all the demand information containing technical terms from the text conversion results, these demand information are combined to determine the demand conversion result. For example, all the information extracted from the text conversion results in the financial field, such as "the annual rate of return needs to be increased" and "the capital turnover cycle needs to be shortened", are combined to form the demand conversion result of the multimodal vertical demand corresponding to the text to be converted.

[0061] The following example from a blasting scenario illustrates the solution provided in this application. The multimodal vertical requirements include the blasting engineer's voice command "Increase the charge, reduce the hole spacing," the on-site recorded text "Too many flying rocks, need to control the range," and the audio extracted from the monitoring video "Too much vibration, check parameters." First, these multimodal requirements are transcribed into text to obtain the text to be converted. From this, the requirement terms such as "charge," "hole spacing," "flying rocks," and "vibration" are extracted, and the terminology vectors for these requirements are determined. Calculations show that these terms have the smallest distance to the "blasting engineering" vertical terminology vector library, thus the vertical category is determined to be blasting engineering. The terminology vectors are input into the corresponding terminology normalization model to obtain initial technical terms, such as "charge" corresponding to "charge amount" and "explosive usage," and "hole spacing" corresponding to "drill hole spacing," etc. Statistics show that "flying rocks" appears most frequently, has the highest weight, and exceeds the preset value; therefore, "blasting debris," with the highest first confidence level, is selected as the technical term. "Vibration" has a lower weight; considering both the first and second confidence levels, "blasting vibration" is selected. The final conversion result is "increased charge, reduced borehole spacing, controlled blast debris range, and checked blast vibration parameters", achieving accurate conversion of blasting scenario requirement terminology into technical terminology.

[0062] By extracting multiple requirement terms from the text to be converted, and combining the word vectors of the requirement terms with the distance between the word vectors of each term in the term vector library of each vertical category and the word vector of the requirement terms, the vertical category to which the text to be converted belongs is determined. Then, the requirement terms in the text to be converted are converted using the terminology normalization model of the corresponding category, and the technical terms corresponding to each requirement term are obtained. The terminology normalization model is determined according to the vertical category, which can accurately convert requirement terms according to different vertical categories, ensuring the accuracy of the conversion results. Then, the requirement terms in the text to be converted are replaced with technical terms to obtain the text conversion result. Finally, the requirement information containing technical terms is extracted from the text conversion result. This can solve the communication problems caused by the difference in terminology between technical personnel and business personnel in vertical industries, and improve communication efficiency and accuracy.

[0063] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0064] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0065] Figure 4 The diagram shows a schematic of the vertical demand conversion device based on multimodal terminology normalization provided in an embodiment of the present invention. For ease of explanation, only the parts related to the embodiment of the present invention are shown, and are described in detail below:

[0066] like Figure 4 As shown, the vertical demand transformation device 4 based on multimodal terminology normalization includes:

[0067] Extraction module 41 is used to extract multiple requirement terms from the text to be converted and determine the corresponding requirement term word vector for each requirement term; the text to be converted is obtained by textualizing multimodal vertical requirements;

[0068] The category determination module 42 is used to determine the vertical category of each text to be converted based on the demand term word vectors of multiple demand terms for each text to be converted;

[0069] The conversion module 43 is used to input the word vector of the demand term of each text to be converted into the term normalization model corresponding to the vertical category of the text to be converted, so as to obtain the technical term corresponding to each demand term.

[0070] The result determination module 44 is used to determine the requirement transformation result of multimodal vertical requirements based on the corresponding technical terms of each requirement term.

[0071] In one possible implementation, the extraction module 41 specifically includes: for each vertical category, training a deep learning model based on the demand term word vectors of demand terms, the annotation results of each demand term, and the technical term word vectors of each technical term in each term sample of the corresponding vertical category, to obtain the term normalization model corresponding to the vertical category; wherein, the term samples are manually annotated term samples; and the annotation results of each demand term are the corresponding technical terms for each demand term.

[0072] In one possible implementation, the extraction module 41 further includes: inputting the voice requirements in the multimodal vertical requirements into a preset speech recognition model to obtain the text to be converted corresponding to the voice requirements; determining the text requirements in the multimodal vertical requirements as the text to be converted corresponding to the text requirements; extracting audio from the video requirements in the multimodal vertical requirements to obtain the audio corresponding to the video requirements, and inputting the audio into a preset speech recognition model to obtain the text to be converted corresponding to the video requirements.

[0073] In one possible implementation, the category determination module 42 specifically includes: for each text to be converted, performing the following steps: calculating the distance between each term vector in the term vector library of the first vertical category and the first demand term word vector of the first demand term in the text to be converted, and determining the minimum value among the distances as the first distance value between the first demand term and the first vertical category; wherein, the first vertical category is any vertical category; the first demand term is any demand term; adding the first distance values ​​of the multiple first demand terms of the text to be converted to the first vertical category to obtain the second distance value between the text to be converted and the first vertical category; and determining the vertical category with the smallest second distance value with the text to be converted as the vertical category of the text to be converted.

[0074] In one possible implementation, the conversion module 43 specifically includes: inputting the word vector of the demand term in each text to be converted into the terminology normalization model corresponding to the vertical category of the text to be converted, to obtain at least one initial technical term for each demand term; determining the frequency of each demand term in the text to be converted, and sorting each demand term in descending order of frequency; determining the weight of each demand term based on the sorting result; and determining the corresponding technical term for each demand term based on the weight of each demand term and the at least one initial technical term corresponding to each demand term.

[0075] In one possible implementation, the conversion module 43 further includes: if a first demand term has a corresponding initial technical term, then the initial technical term is determined as the technical term of the first demand term; wherein, the first demand term is any demand term; if the first demand term has multiple corresponding initial technical terms, then the following steps are performed: determining a first confidence level for each technical term based on the distance between the demand term word vector of the first demand term and the demand term word vector of each initial technical term; determining a second confidence level for each technical term based on the semantic matching degree of each initial technical term of the first demand term; and determining the technical term of the first demand term based on the weight of each first demand term, the first confidence level and the second confidence level of each initial technical term corresponding to the first demand term.

[0076] In one possible implementation, the conversion module 43 further includes: if the weight of the first demand term exceeds a preset weight, then the initial technical term with the highest first confidence level is determined as the technical term of the first demand term; if the weight of the first demand term does not exceed the preset weight, then the sum of the first confidence level and the second confidence level of each technical term is determined as the third confidence level, and the initial technical term with the highest third confidence level is determined as the technical term of the first demand term.

[0077] In one possible implementation, the result determination module 42 specifically includes: replacing each requirement term in the text to be converted with the corresponding technical term to obtain the text conversion result of the text to be converted; extracting multiple requirement information containing technical terms from the text conversion result, and determining all requirement information of the sample to be converted as the requirement conversion result of the corresponding multimodal vertical requirement.

[0078] Figure 5 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 5 As shown, the electronic device 5 of this embodiment includes a processor 50 and a memory 51. The memory 51 stores a computer program 52. When the processor 50 executes the computer program 52, it implements the steps in the various method embodiments described above. Alternatively, when the processor 50 executes the computer program 52, it implements the functions of each module / unit in the various device embodiments described above.

[0079] For example, computer program 52 may be divided into one or more modules / units, which are stored in memory 51 and executed by processor 50 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 52 in electronic device 5.

[0080] Electronic device 5 may include, but is not limited to, processor 50 and memory 51. Those skilled in the art will understand that... Figure 5 This is merely an example of electronic device 5 and does not constitute a limitation on electronic device 5. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 5 may also include input / output devices, network access devices, buses, etc.

[0081] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0082] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0083] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for transforming vertical requirements based on multimodal terminology normalization, characterized in that, include: Extract multiple requirement terms from the text to be converted, and determine the corresponding requirement term word vector for each requirement term; The text to be converted is obtained by textualizing the multimodal vertical requirements; Based on the demand term word vectors of multiple demand terms for each text to be converted, determine the vertical category of each text to be converted; Input the term vector of the demand term for each text to be converted into the term normalization model corresponding to the vertical category of the text to be converted to obtain the corresponding technical term for each demand term. Based on the corresponding technical terms of each requirement term, the requirement transformation result of the multimodal vertical requirement is determined; The step involves inputting the word vector of the demand term for each text to be converted into the terminology normalization model corresponding to the vertical category of the text to be converted, thereby obtaining the corresponding technical terms for each demand term, including: Input the term vector of each demand term in the text to be converted into the term normalization model corresponding to the vertical category of the text to be converted, and obtain at least one initial technical term for each demand term; Determine the frequency of each requirement term in the text to be converted, and sort each requirement term in descending order of frequency. Based on the ranking results of each requirement term, determine the weight of each requirement term; Based on the weight of each demand term and at least one initial technical term corresponding to each demand term, the corresponding technical term is determined.

2. The method for vertical demand transformation based on multimodal terminology normalization according to claim 1, characterized in that, The process of determining the technical terms for each requirement term based on its weight and at least one initial technical term includes: If a corresponding initial technical term exists for the first requirement term, then the initial technical term is determined as the technical term for the first requirement term; wherein, the first requirement term is any requirement term. If there are multiple corresponding initial technical terms for the first requirement term, then perform the following steps: Based on the distance between the word vector of the first demand term and the word vector of the demand term of each initial technical term, the first confidence level of each technical term is determined; Based on the semantic matching degree of each initial technical term of the first demand term, determine the second confidence level of each technical term; The technical terminology of the first requirement term is determined based on the weight of each first requirement term and the first and second confidence levels of each initial technical term corresponding to the first requirement term.

3. The method for vertical demand transformation based on multimodal terminology normalization according to claim 2, characterized in that, The process of determining the technical terms of the first requirement term based on the weight of each first requirement term and the first and second confidence levels of each corresponding technical term includes: If the weight of the first requirement term exceeds the preset weight, then the initial technical term with the highest confidence level is determined as the technical term of the first requirement term. If the weight of the first requirement term does not exceed the preset weight, then the sum of the first confidence level and the second confidence level of each technical term is determined as the third confidence level, and the initial technical term with the highest third confidence level is determined as the technical term of the first requirement term.

4. The method for vertical demand transformation based on multimodal terminology normalization according to claim 1, characterized in that, The determination of the demand transformation result of the multimodal vertical demand based on the corresponding technical terms of each demand term includes: Replace each requirement term in the text to be converted with the corresponding technical term to obtain the text conversion result of the text to be converted; From the text conversion results, multiple pieces of demand information containing technical terms are extracted, and all the demand information of the sample to be converted are determined as the corresponding multimodal vertical demand conversion results.

5. The method for vertical demand transformation based on multimodal terminology normalization according to claim 1, characterized in that, The process of determining the vertical category of each text to be converted based on the demand term word vectors of multiple demand terms for each text to be converted includes: For each text to be converted, perform the following steps: Calculate the distance between each term vector in the term vector library of the first vertical category and the first demand term word vector in the text to be converted, and determine the minimum value among the distances as the first distance value between the first demand term and the first vertical category; wherein, the first vertical category is any vertical category; and the first demand term is any demand term; Add the first distance value of the first requirement terms of the text to be converted to the first vertical category to obtain the second distance value between the text to be converted and the first vertical category; The vertical category with the smallest second distance value to the text to be converted is determined as the vertical category of the text to be converted.

6. The method for vertical demand transformation based on multimodal terminology normalization according to claim 1, characterized in that, The textual processing of multimodal vertical requirements includes: The voice requirements from the multimodal vertical requirements are input into a preset speech recognition model to obtain the corresponding text to be converted. The text requirements in the multimodal vertical category requirements are identified as the text to be converted corresponding to the text requirements; Audio is extracted from the video requirements in the multimodal vertical category to obtain the corresponding audio for the video requirements. The audio is then input into a preset speech recognition model to obtain the corresponding text to be converted for the video requirements.

7. The method for vertical demand transformation based on multimodal terminology normalization according to claim 1, characterized in that, Before extracting multiple demand terms from the text to be converted and determining the corresponding demand term word vector for each demand term, the method further includes: For each vertical category, a deep learning model is trained based on the demand term word vectors of demand terms, the annotation results of each demand term, and the technical term word vectors of each technical term in each term sample of the corresponding vertical category, to obtain the term normalization model corresponding to the vertical category; wherein, the term samples are manually annotated term samples; and the annotation results of each demand term are the corresponding technical terms.

8. A vertical demand transformation device based on multimodal terminology normalization, characterized in that, include: The extraction module is used to extract multiple demand terms from the text to be converted and determine the corresponding demand term word vector for each demand term. The text to be converted is obtained by textualizing the multimodal vertical requirements; The category determination module is used to determine the vertical category of each text to be converted based on the demand term word vectors of multiple demand terms for each text to be converted; The conversion module is used to input the word vector of the demand term in each text to be converted into the terminology normalization model corresponding to the vertical category of the text to be converted, so as to obtain the corresponding technical terminology for each demand term. The result determination module is used to determine the requirement transformation result of the multimodal vertical requirement based on the corresponding technical terms of each requirement term. The conversion module is also used to input the demand term word vector of each text to be converted into the term normalization model corresponding to the vertical category of the text to be converted, so as to obtain at least one initial technical term for each demand term. Determine the frequency of each requirement term in the text to be converted, and sort each requirement term in descending order of frequency; determine the weight of each requirement term based on the sorting result. Based on the weight of each demand term and at least one initial technical term corresponding to each demand term, the corresponding technical term is determined.

9. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 7.

Images