AI analysis device for interrogation intention recognition

Abstract

The invention relates to the field of online interrogation, in particular to an AI analysis device for interrogation intention recognition, which is characterized in that an interrogation AI model is utilized to generate various answer texts for questions of a user, and then an intention entity and an emotion entity are extracted from a dialogue text of the current round of the user; therefore, intention changes of users in the current round of dialogue and the previous round of dialogue are analyzed, and the intention changes comprise emotion changes. The answer text most conforming to the current intention of the user is selected based on the intention change, so that the finally output answer text can better reflect the intention of the user, and the output answer is prevented from deviating from the expectation of the user.

Description

Technical Field

[0001] The present invention relates to the field of online medical consultations, and specifically to an AI analysis device for identifying consultation intentions. Background Art

[0002] Online medical consultation, as an important part of digital health services, has witnessed rapid development and popularity in recent years. It connects medical service providers and patients through Internet technology, providing various conveniences and additional values.

[0003] Existing online medical consultation technologies are generally built based on AI models. The AI model is similar to a black box model, which can analyze preliminary conclusions (such as disease labels) based on the symptom information input by users, so as to match appropriate medical resources according to the conclusions for further diagnosis and medical assistance, etc.

[0004] The current difficulty of online medical consultation technology is that it is difficult to understand the intention changes of users during the multi-round conversation process, which may lead to the replies output by the AI model not meeting the needs of users. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide an AI analysis device for identifying consultation intentions to solve the problems in the background art.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] An AI analysis device for identifying consultation intentions of the present invention includes:

[0008] An acquisition module for acquiring the conversation text of the current conversation round from the user;

[0009] A answering module for taking the conversation text of the current conversation round as input, invoking the consultation AI conversation model, and obtaining multiple answering texts of the current conversation round;

[0010] An entity extraction module for performing target entity extraction on the conversation text of the current conversation round to obtain the target entity of the current conversation round, where the target entity includes a symptom entity, an intention entity, and an emotion entity;

[0011] An intention analysis module for analyzing the intention change result of the user based on the target entity of the current conversation round and the target entity of the previous conversation round, and selecting a target answering text from the multiple answering texts of the current conversation round based on the intention change result.

[0012] In an embodiment of the present application, taking the conversation text of the current conversation round as input, invoking the consultation AI conversation model, and obtaining multiple answering texts of the current conversation round includes:

[0013] Tokenize the dialogue text of the current dialogue turn to obtain multiple words of the current dialogue turn;

[0014] Vectorize the multiple words of the current dialogue turn to obtain a sequence of word vectors, where the word vectors include semantic encoding and position encoding;

[0015] Construct an input vector based on the sequence of word vectors and input the input vector into the retrieved AI dialogue model for medical consultation to obtain multiple response texts for the current dialogue turn.

[0016] In an embodiment of the present application, extracting target entities from the dialogue text of the current dialogue turn to obtain the target entities of the current dialogue turn includes:

[0017] Match the semantic encoding of each word vector in the sequence of word vectors with a pre-constructed template vector to obtain a matching degree;

[0018] Use the word vectors with a matching degree greater than a preset threshold as target entities, and the entity type of the word vector is consistent with the entity type of the template vector.

[0019] In an embodiment of the present application, analyzing the result of the user's intention change based on the target entities of the current dialogue turn and the target entities of the previous dialogue turn includes:

[0020] Construct the medical consultation intention of the current dialogue turn based on symptom entities and intention entities; calculate the similarity between the medical consultation intention of the current dialogue turn and the medical consultation intention of the previous dialogue turn; calculate the emotion score of the current dialogue turn based on emotion entities, and calculate the emotion change feature based on the emotion score of the current dialogue turn and the emotion score of the previous dialogue turn;

[0021] Analyze the result of the user's intention change based on the similarity between the medical consultation intention of the current dialogue turn and the medical consultation intention of the previous dialogue turn and the emotion change feature.

[0022] In an embodiment of the present application, analyzing the result of the user's intention change based on the similarity between the medical consultation intention of the current dialogue turn and the medical consultation intention of the previous dialogue turn and the emotion change feature includes:

[0023] When the similarity between the medical consultation intention of the current dialogue turn and the medical consultation intention of the previous dialogue turn is less than a preset similarity threshold, determine that the result of the intention change is a switch of the medical consultation intention;

[0024] When the similarity between the medical consultation intention of the current dialogue turn and the medical consultation intention of the previous dialogue turn is greater than or equal to the preset similarity threshold and the emotion change feature is a positive change, determine that the result of the intention change is an intention deepening;

[0025] When the similarity between the interrogation intention in the current conversation turn and the interrogation intention in the previous conversation turn is greater than or equal to a preset similarity threshold and the emotional change feature is no change, it is determined that the intention change result is intention maintenance;

[0026] When the similarity between the interrogation intention in the current conversation turn and the interrogation intention in the previous conversation turn is greater than or equal to a preset similarity threshold and the emotional change feature is negative change, it is determined that the intention change result is intention deviation.

[0027] In an embodiment of the present application, an emotional score of the current conversation turn is calculated based on emotional entities, and an emotional change feature is calculated based on the emotional score of the current conversation turn and the emotional score of the previous conversation turn, including:

[0028] Emotionally annotate the emotional entities based on a pre-constructed emotion annotation model to obtain the emotion scores of each emotional entity;

[0029] Calculate the average value of the emotion scores of all emotional entities in the conversation text of the current conversation turn to obtain the emotional score of the current conversation turn;

[0030] Calculate the difference between the emotional score of the current conversation turn and the emotional score of the previous conversation turn to obtain the emotional change feature.

[0031] In an embodiment of the present application, a target answer text is selected from multiple answer texts of the current conversation turn based on the intention change feature, including:

[0032] When the intention change result is interrogation intention switching, the answer text with the highest confidence in the multiple answer texts of the current conversation turn is used as the target answer text.

[0033] In an embodiment of the present application, a target answer text is selected from multiple answer texts of the current conversation turn based on the intention change feature, including:

[0034] When the intention change result is intention deepening or intention maintenance, extract the topic words of the answer text in the previous conversation turn and the topic words of the multiple answer texts in the current conversation turn;

[0035] Calculate the similarity between the topic words of the answer text in the previous conversation turn and the topic words of the multiple answer texts in the current conversation turn, and calculate the content depth of the multiple answer texts in the current conversation turn, where the content depth is determined based on the number of words and word diversity of the answer text;

[0036] Perform a weighted sum of the similarity and content depth of the topic words of each response text in the current conversation turn to obtain a comprehensive evaluation score, and select the response text with the highest comprehensive evaluation score as the target response text.

[0037] In an embodiment of the present application, when the intention change result is intention deviation, extract the topic words of the response text in the previous conversation turn and the topic words of multiple response texts in the current conversation turn;

[0038] Calculate the similarity between the topic words of the response text in the previous conversation turn and the topic words of multiple response texts in the current conversation turn, and determine the target response text based on the value of the emotion change feature and the similarity. Among them, the smaller the value of the emotion change feature, the smaller the similarity of the selected response text.

[0039] In an embodiment of the present application, the intention analysis module is further configured to, when no target entity is extracted from the conversation text of the current conversation turn, use the response text with the highest confidence in multiple response texts of the current conversation turn as the target response text.

[0040] The beneficial effects of the present invention are as follows: An AI analysis device for medical consultation intention recognition according to the present invention. The present application uses a medical consultation AI model to generate multiple response texts for a user's question, and then extracts intention entities and emotion entities from the conversation text of the user's current turn, so as to analyze the intention change of the current turn of the conversation compared with the user's intention in the previous conversation turn. The intention change includes emotion change. Select the response text that best conforms to the user's current intention based on the intention change, so that the finally output response text can better reflect the user's intention and avoid the output answer deviating from the user's expectation. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention will be further described below in conjunction with the drawings and embodiments:

[0042] Figure 1 is a structural diagram of an AI analysis device for medical consultation intention recognition shown in an embodiment of the present application;

[0043] Figure 2 is a schematic flow chart of multi-text selection based on intention analysis in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand the other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0045] It should be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner. Therefore, only the layers related to the present invention are shown in the diagrams, rather than being drawn according to the number, shape, and size of the layers in actual implementation. The type, quantity, and ratio of each layer in actual implementation can be arbitrarily changed, and the layer layout type may also be more complex.

[0046] In the following description, a large number of details are explored to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details.

[0047] Figure 1 It is a structural diagram of an AI analysis device for identifying the intention of medical consultation shown in an embodiment of the present application. As Figure 1 shown: An AI analysis device for identifying the intention of medical consultation in this embodiment includes:

[0048] An acquisition module 110, configured to acquire the conversation text of the current conversation turn from the user;

[0049] In the present application, an online medical consultation is realized by adopting a multi-turn conversation method.

[0050] During the multi-turn conversation process, the user will input question text to obtain the result they want. If the text replied by the system is inconsistent with the expectation, then the user will ask again, thus conducting a multi-turn conversation.

[0051] A response module 120, configured to take the conversation text of the current conversation turn as input, retrieve the medical consultation AI conversation model, and obtain multiple response texts for the current conversation turn;

[0052] In the present application, based on the large language model and the existing medical consultation conversation model to handle the question text input by the user. In the present application, the obtained response texts will not be directly output to the user, but are retained in the system. Among them, if multiple response texts are output, the confidence of each response text will also be retained.

[0053] For example, the question text is, "I have always felt fatigued recently and occasionally have symptoms of dizziness. May I ask what might be the cause?"

[0054] Based on this input, the AI model may generate the following multiple response texts and their confidence levels:

[0055] Answer 1: Text: "The symptoms you described may be related to anemia, which can cause the body's tissues to not receive enough oxygen, leading to fatigue and dizziness."

[0056] Confidence level: 0.75

[0057] Answer 2: Text: "Your symptoms may also be caused by lack of sleep or poor sleep quality. Chronic sleep deprivation can affect physical health, causing fatigue and inattention."

[0058] Confidence level: 0.68

[0059] Answer 3: Text: "These symptoms may also be related to hypothyroidism, which reduces the metabolic rate, leading to a decrease in energy levels and feelings of fatigue."

[0060] Confidence level: 0.62

[0061] Answer 4: Text: "Sometimes, excessive stress or anxiety can also trigger similar symptoms. It is also important to consider the impact of psychological factors on the body."

[0062] Confidence level: 0.55

[0063] Specifically, in this application, the text is converted into a vector and input into the model in the following manner, including:

[0064] S210, perform word segmentation on the dialogue text of the current dialogue turn to obtain multiple words of the current dialogue turn;

[0065] The word segmentation algorithm uses the existing bidirectional maximum matching.

[0066] S220, perform vectorization processing on the multiple words of the current dialogue turn to obtain a sequence of word vectors, where the word vectors include semantic encoding and position encoding;

[0067] In this application, when performing vectorization processing on the word sequence, during vectorization, a specific position encoding mechanism included in the Transformer architecture is used to ensure that the model can identify the position of the vocabulary in the sequence.

[0068] S230, construct an input vector based on the sequence of word vectors and input the input vector into the retrieved medical consultation AI dialogue model to obtain multiple response texts of the current dialogue turn.

[0069] Finally, the word vector sequence with semantic encoding and position encoding retained is used as the input vector and input into the consultation AI dialogue model, and multiple response texts for the current dialogue turn can be obtained.

[0070] The entity extraction module 130 is used to extract target entities from the dialogue text of the current dialogue turn to obtain the target entities of the current dialogue turn, where the target entities include symptom entities, intention entities, and emotion entities.

[0071] Among them, the target entities include symptom entities, intention entities, and emotion entities. The intention entity can be a relationship entity, a request entity, etc. For example, for "What is the cause of dizziness", "the cause is" is a relationship entity.

[0072] The process of extracting target entities includes:

[0073] S310: Match the semantic encoding of each word vector in the word vector sequence with a pre-constructed template vector to obtain a matching degree.

[0074] S320: Use the word vectors with a matching degree greater than a preset threshold as target entities, and the entity type of the word vector is the same as the entity type of the template vector.

[0075] Among them, the matching degree is the cosine similarity.

[0076] The intention analysis module 140 is used to analyze the result of the user's intention change based on the target entities of the current dialogue turn and the target entities of the previous dialogue turn, and select a target response text from the multiple response texts of the current dialogue turn based on the result of the intention change. When no target entity is extracted from the dialogue text of the current dialogue turn, the response text with the highest confidence in the multiple response texts of the current dialogue turn is used as the target response text.

[0077] This application performs intention analysis based on the above entities. Figure 2 This is a schematic diagram of the multi-text selection process based on intention analysis in an embodiment of this application. As Figure 2 shown, specifically, it mainly reflects whether the previous response meets the user's expectations based on the user's intention change and emotion change, and selects multiple response texts based on the result of the intention change. The specific process is as follows:

[0078] (1) Analysis of the result of intention change

[0079] (1-1) Construct the consultation intention of the current dialogue turn based on the symptom entity and the intention entity; calculate the similarity between the consultation intention of the current dialogue turn and the consultation intention of the previous dialogue turn; calculate the emotion score of the current dialogue turn based on the emotion entity, and calculate the emotion change feature based on the emotion score of the current dialogue turn and the emotion score of the previous dialogue turn.

[0080] Among them, this application mainly analyzes the intention of the medical interview. Therefore, it is necessary to combine symptom entities. Using symptom entities and relationship entities, a template structure similar to two elements of a triple can be combined. For example, what is the cause of coughing, and "coughing" + "the cause is" constitutes the intention of the medical interview.

[0081] The change in the intention of the medical interview in two rounds of conversations can directly reflect the user's inquiry intention. This application uses similarity calculation to determine the change in the intention of the medical interview in two rounds of conversations. If the change in the intention of the current round is too large compared to the previous round of conversation, it indicates that the user is shifting the inquiry intention. At this time, the conversation relationship between the current round and the previous round can be separated. If there is similarity, it means that the user is asking questions based on the previous round of conversation. And the emotional change feature is one of the criteria for judging whether the answer text in the previous round meets the user's expectations.

[0082] The methods for extracting emotional change features include:

[0083] (1-1-1) Perform emotional annotation on emotional entities based on a pre-constructed emotional annotation model to obtain the emotional scores of each emotional entity;

[0084] In this application, a dataset with emotional labels is used to train the model. These datasets usually contain text and its corresponding emotional categories (such as happy, sad, angry, etc.) or emotional scores (such as continuous values within the range of [-1,1] or [0,1]). The model can be used to quickly annotate the emotional scores of emotional entities.

[0085] (1-1-2) Calculate the average value of the emotional scores of all emotional entities in the conversation text of the current conversation round to obtain the emotional score M of the current conversation round n , where n is the current round;

[0086] (1-1-3) Calculate the difference between the emotional score of the current conversation round and the emotional score of the previous conversation round to obtain the emotional change feature.

[0087] The value of the emotional change feature CM = M n -M n-1 , since the emotional score of each round is between [0,1], the calculated score is between [-1,1].

[0088] (1-2) Analyze the result of the user's intention change based on the similarity between the intention of the current conversation round and the intention of the previous conversation round and the emotional change feature.

[0089] There are the following four situations for the result of the intention change and the corresponding selection of the answer text:

[0090] (a) When the similarity between the inquiry intention in the current conversation turn and the inquiry intention in the previous conversation turn is less than the preset similarity threshold, it is determined that the intention change result is a switch in the inquiry intention;

[0091] In this application, a similarity threshold is set according to the requirements of the actual application scenario. If the calculated similarity is less than this threshold, it is determined that the inquiry intention has switched.

[0092] For example: In the previous conversation: "Why have I been feeling tired recently and occasionally dizzy?" The inquiry intentions are "tiredness", "dizziness", "why"

[0093] "I also have back pain recently. How can I improve it?" The inquiry intentions are "back pain" and "improve". Although both involve discussions about fatigue symptoms, there is not much similarity between them, and the inquiry intention is generally in a switched state.

[0094] At this time, regardless of the user's intention change, the response text with the highest confidence in the multiple response texts of the current conversation turn is used as the target response text.

[0095] (b) When the similarity between the inquiry intention in the current conversation turn and the inquiry intention in the previous conversation turn is greater than or equal to the preset similarity threshold, and the emotional change feature is a positive change, it is determined that the intention change result is an intention deepening;

[0096] For example:

[0097] In the previous conversation: "I have been feeling tired recently and occasionally dizzy." The inquiry intentions are "tiredness", "dizziness", "why".

[0098] Answer in the previous conversation: "You may not be getting enough rest, or please confirm if you have a cold."

[0099] In the current conversation turn: "Okay, then how can I improve my fatigue and dizziness?" The inquiry intentions are "tiredness", "dizziness", "improve".

[0100] This sentence still focuses on the previous symptoms, but more specifically asks how to improve. The emotional analysis shows that the emotion has slightly improved, manifested as a positive attitude change from worry to seeking solutions. This also indicates the user's recognition of the previous answer. Therefore, when determining the response text for the current turn, the text of the previous turn's answer can be used as a reference, and a text with a deepened response content can be found. Specifically including:

[0101] (b-1) When the intention change result is intention deepening or intention maintaining, extract the topic words of the response text in the previous conversation turn and the topic words of the multiple response texts in the current conversation turn;

[0102] Use a large language model to summarize the topic words of the answer text in the previous conversation turn and the topic words of multiple answer texts in the current conversation turn. For example: after the input text, enter "Please help me summarize the topic words of the above text." in the large language model.

[0103] (b-2) Calculate the similarity between the topic words of the answer text in the previous conversation turn and the topic words of multiple answer texts in the current conversation turn, and calculate the content depth of multiple answer texts in the current conversation turn, where the content depth is determined based on the number of words and word diversity of the answer text;

[0104] The calculation of similarity still uses vectorization and the method of calculating cosine similarity. Since the user approves of the answer text in the previous conversation turn and the question raised in this turn is similar to the question in the previous turn, it can be considered that the user is asking a follow-up question based on the answer text in the previous turn. Therefore, it is only necessary to find the content that is relevant to the answer text in the previous turn and has a deeper content from multiple candidate answer texts in this turn.

[0105] In this application, the content depth D is determined by counting the number of words and word diversity. The mathematical expression of the content depth is:

[0106]

[0107] In the formula, W1 is the first weight, W2 is the second weight, Normalized() represents the normalization function, N is the number of words in this article, N w is the number of word types in the text, represents the quantization index of the word diversity of the text.

[0108] (b-3) Perform a weighted sum of the similarity of the topic words and the content depth of each answer text in the current conversation turn to obtain a comprehensive evaluation score, and select the answer text with the highest comprehensive evaluation score as the target answer text.

[0109] The comprehensive evaluation score C = W3×D + W4×N, where W3 is the third weight and W4 is the fourth weight.

[0110] (c) When the similarity between the medical consultation intention in the current conversation turn and the medical consultation intention in the previous conversation turn is greater than or equal to the preset similarity threshold and the emotion change feature is no change, determine that the intention change result is intention retention;

[0111] Situation (c) is similar to situation (b), so the answer text selection in situation (b) can be used.

[0112] (d) When the similarity between the inquiry intention in the current conversation turn and the inquiry intention in the previous conversation turn is greater than or equal to a preset similarity threshold and the emotional change feature is a negative change, it is determined that the intention change result is intention deviation.

[0113] If negative emotions are extracted from the answer returned by the user, it can be determined that the answer in the previous conversation turn did not make the user recognize it. Therefore, when determining the answer in the current turn, it is necessary to avoid the direction of the answer text in the previous turn and let the topic words deviate based on the negative degree of the emotion. Specifically, it includes:

[0114] (d-1) When the intention change result is intention deviation, extract the topic words of the answer text in the previous conversation turn and the topic words of multiple answer texts in the current conversation turn;

[0115] (d-2) Calculate the similarity between the topic words of the answer text in the previous conversation turn and the topic words of multiple answer texts in the current conversation turn, and determine the target answer text based on the value of the emotional change feature and the similarity. Among them, the smaller the value of the emotional change feature, the smaller the similarity of the selected answer text.

[0116] If the value of the emotional change feature in this application is negative, it means a phase change, and the deviation degree is determined according to the negative change value.

[0117] Specifically, this application sorts multiple answer texts according to similarity, and then delimits the levels of the values of multiple emotional change features. For example, there are 5 levels, each level corresponds to a value range, and 5 answer texts are correspondingly selected. Calculate the level into which the value of the emotional change feature falls, and then select the answer text according to the corresponding level.

[0118] An AI analysis device for inquiry intention recognition according to the present invention. This application uses an inquiry AI model to generate multiple answer texts for the user's question, and then extracts the intention entity and the emotion entity from the conversation text of the user in the current turn, so as to analyze the intention change of the conversation in the current turn and the intention of the user in the previous conversation turn. The intention change includes emotional change. Select the answer text that best conforms to the user's current intention based on the intention change, so that the finally output answer text can better reflect the user's intention and avoid the deviation of the output answer from the user's expectation.

[0119] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements any one of the methods in this embodiment, where the method is the execution logic of this system.

[0120] This embodiment also provides an electronic terminal, including: a processor and a memory;

[0121] The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory, so that the terminal executes any of the methods in this embodiment.

[0122] For the computer-readable storage medium in this embodiment, those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to computer programs. The foregoing computer program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: various media such as ROM, RAM, magnetic disk, or optical disc that can store program codes.

[0123] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication with each other. The memory is used to store computer programs, the communication interface is used for communication, and the processor and the transceiver are used to run the computer programs to make the electronic terminal execute each step of the above method.

[0124] In this embodiment, the memory may include a Random Access Memory (RAM), and may also include a non-volatile memory, such as at least one disk memory.

[0125] The above-mentioned processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0126] In the above embodiments, although the present invention has been described in combination with specific embodiments of the present invention, according to the previous description, many substitutions, modifications, and variations of these embodiments will be obvious to those of ordinary skill in the art. The embodiments of the present invention are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims.

[0127] The above embodiments are only illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical idea disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. An AI analysis device for medical interview intention recognition, characterized in that Including: An acquisition module, configured to acquire the dialogue text of the current dialogue turn from the user; An answering module, configured to use the dialogue text of the current dialogue turn as input, retrieve the medical consultation AI dialogue model, and obtain multiple answering texts for the current dialogue turn; An entity extraction module, configured to perform target entity extraction on the dialogue text of the current dialogue turn to obtain the target entity of the current dialogue turn, where the target entity includes a symptom entity, an intention entity, and an emotion entity; An intention analysis module, configured to analyze the result of the user's intention change based on the target entity of the current dialogue turn and the target entity of the previous dialogue turn, and select a target answering text from the multiple answering texts of the current dialogue turn based on the result of the intention change.

2. The AI analysis device for medical interview intention recognition according to claim 1, characterized in that, Using the dialogue text of the current dialogue turn as input, retrieving the medical consultation AI dialogue model, and obtaining multiple answering texts for the current dialogue turn, including: Performing word segmentation processing on the dialogue text of the current dialogue turn to obtain multiple words of the current dialogue turn; Performing vectorization processing on the multiple words of the current dialogue turn to obtain a word vector sequence, where the word vector includes a semantic encoding and a position encoding; Constructing an input vector based on the word vector sequence, and inputting the input vector into the retrieved medical consultation AI dialogue model to obtain multiple answering texts for the current dialogue turn.

3. The AI analysis device for medical consultation intention recognition according to claim 2, characterized in that, Performing target entity extraction on the dialogue text of the current dialogue turn to obtain the target entity of the current dialogue turn, including: Matching the semantic encoding of each word vector in the word vector sequence with a pre-constructed template vector to obtain a matching degree; Taking the word vectors with a matching degree greater than a preset threshold as target entities, and the entity type of the word vector is consistent with the entity type of the template vector.

4. The AI analysis device for medical interview intention recognition according to claim 2, wherein Analyzing the result of the user's intention change based on the target entity of the current dialogue turn and the target entity of the previous dialogue turn, including: Constructing the medical consultation intention of the current dialogue turn based on the symptom entity and the intention entity; calculating the similarity between the medical consultation intention of the current dialogue turn and the medical consultation intention of the previous dialogue turn; calculating the emotion score of the current dialogue turn based on the emotion entity, and calculating the emotion change feature based on the emotion score of the current dialogue turn and the emotion score of the previous dialogue turn; Analyzing the result of the user's intention change based on the similarity between the medical consultation intention of the current dialogue turn and the medical consultation intention of the previous dialogue turn and the emotion change feature.

5. An AI analysis device for medical interview intention recognition according to claim 4, characterized in that, Analyzing the result of the user's intention change based on the similarity between the medical consultation intention of the current dialogue turn and the medical consultation intention of the previous dialogue turn and the emotion change feature, including: When the similarity between the medical consultation intention of the current dialogue turn and the medical consultation intention of the previous dialogue turn is less than a preset similarity threshold, determining that the result of the intention change is a medical consultation intention switch; When the similarity between the medical consultation intention of the current dialogue turn and the medical consultation intention of the previous dialogue turn is greater than or equal to the preset similarity threshold, and the emotion change feature is a positive change, determining that the result of the intention change is an intention deepening; When the similarity between the inquiry intention in the current dialogue turn and the inquiry intention in the previous dialogue turn is greater than or equal to a preset similarity threshold, and the emotional change feature is no change, the intention change result is determined to be intention maintenance; When the similarity between the inquiry intention in the current dialogue turn and the inquiry intention in the previous dialogue turn is greater than or equal to a preset similarity threshold, and the emotional change feature is a negative change, the intention change result is determined to be intention deviation.

6. An AI analysis device for medical interview intention recognition according to claim 4, characterized in that, Calculate the emotional score of the current dialogue turn based on the emotional entity, and calculate the emotional change feature based on the emotional score of the current dialogue turn and the emotional score of the previous dialogue turn, including: Perform emotional annotation on the emotional entity based on a pre-constructed emotional annotation model to obtain the emotional score of each emotional entity; Calculate the average value of the emotional scores of all emotional entities in the dialogue text of the current dialogue turn to obtain the emotional score of the current dialogue turn; Calculate the difference between the emotional score of the current dialogue turn and the emotional score of the previous dialogue turn to obtain the emotional change feature.

7. An AI analysis device for medical interview intention recognition according to claim 5, characterized in that, Select the target response text from multiple response texts in the current dialogue turn based on the intention change feature, including: When the intention change result is an inquiry intention switch, use the response text with the highest confidence in the multiple response texts of the current dialogue turn as the target response text.

8. An AI analysis device for medical interview intention recognition according to claim 5, characterized in that, Select the target response text from multiple response texts in the current dialogue turn based on the intention change feature, including: When the intention change result is intention deepening or intention maintenance, extract the topic words of the response text in the previous dialogue turn and the topic words of the multiple response texts in the current dialogue turn; Calculate the similarity between the topic words of the response text in the previous dialogue turn and the topic words of the multiple response texts in the current dialogue turn, and calculate the content depth of the multiple response texts in the current dialogue turn, where the content depth is determined based on the number of words and word diversity of the response text; Perform a weighted sum of the similarity of the topic words and the content depth of each response text in the current dialogue turn to obtain a comprehensive evaluation score, and select the response text with the highest comprehensive evaluation score as the target response text.

9. An AI analysis device for identifying medical consultation intentions according to claim 5, characterized in that When the intention change result is intention deviation, extract the topic words of the response text in the previous dialogue turn and the topic words of the multiple response texts in the current dialogue turn; Calculate the similarity between the topic words of the response text in the previous dialogue turn and the topic words of the multiple response texts in the current dialogue turn, and determine the target response text based on the value of the emotional change feature and the similarity, where the smaller the value of the emotional change feature, the smaller the similarity of the selected response text.

10. An AI analysis device for medical interview intention recognition according to claim 1, characterized in that, The intention analysis module is also used to use the response text with the highest confidence in the multiple response texts of the current dialogue turn as the target response text when no target entity is extracted from the dialogue text of the current dialogue turn.

Images