Medical equipment interaction method and device, storage medium and medical equipment

Abstract

The invention discloses an interaction method and device of medical equipment, a storage medium and the medical equipment. The method comprises the following steps: in response to voice information received by medical equipment, converting the voice information into a voice text; inputting the voice text into a target recognition model to obtain a target personality feature possessed by an output person of the voice information; obtaining a feedback text and / or a feedback action matched with the voice text; updating vocabularies of the feedback text and / or control parameters of the feedback action according to the target personality characteristics; and outputting the updated feedback information and / or executing a feedback action. According to the method, automatic personality prediction is achieved, meanwhile, the influence caused by missing of the text data set can be made up, and the personality prediction effect is improved. Besides, the medical equipment can make corresponding emotional response according to the unique personality characteristics embodied in the user voice information, emotional communication between the medical equipment and the user is increased, and individuation and high efficiency of service are ensured.

Description

Technical Field

[0001] The present application relates to the field of smart medical technology, and in particular to an interaction method, device, storage medium and medical device for medical equipment. Background Art

[0002] The development of large language models (e.g., ChatGPT) has led to a wide range of applications in human-computer interaction (HCI), including AI-based psychotherapy, educational robotics, and companion robots for the elderly.

[0003] In related technologies, medical robots often ignore implicit factors in the conversation, such as the patient's personality and emotions, which causes the medical robot's feedback to focus only on the input conversation content and ignore the semantic understanding of the personality itself. This leads to a lack of emotional communication with the user, making it difficult to achieve real communication and unable to truly "prescribe the right medicine", which reduces the quality of service.

[0004] Therefore, there is an urgent need for a medical robot that can capture multi-dimensional information and emotional fluctuations in conversations and provide a deep understanding of individual personality traits. Summary of the invention

[0005] In view of this, the present application provides an interactive method, apparatus, storage medium and medical device for a medical device, which can provide explainable interactive results by linking the language and behavior of the voice outputter in the conversation with the description of personality characteristics, thereby ensuring the personalization and high efficiency of the service.

[0006] According to one aspect of the present application, a medical device interaction method is provided, comprising:

[0007] In response to the medical device receiving the voice information, converting the voice information into voice text;

[0008] Inputting the speech text into a target recognition model to obtain the target personality characteristics of the outputter of the speech information, wherein the target recognition model is obtained by fine-tuning a pre-trained language model through the confidence probabilities of the interlocutor having and not having the personality characteristics in the historical conversation text;

[0009] Acquiring feedback text and / or feedback action matching the voice text;

[0010] updating the vocabulary of the feedback text and / or the control parameters of the feedback action according to the target personality characteristics;

[0011] Output the updated feedback information and / or execute the feedback action.

[0012] Optionally, the interaction method of the medical device further includes:

[0013] Acquire the historical conversation text, wherein the historical conversation text carries an emotion tag and an identity identifier of the interlocutor;

[0014] Identifying personality characteristics of different interlocutors in the historical conversation text based on the emotion tag and the identity identifier;

[0015] Constructing a control text without the individual characteristics based on the historical dialogue text;

[0016] The historical conversation text and the control text are used as positive and negative samples respectively, and the pre-trained language model is fine-tuned to obtain a target recognition model.

[0017] Optionally, identifying personality characteristics of different interlocutors in the historical conversation text based on the emotion tag and the identity identifier includes:

[0018] Determining personality information of the historical dialogue text according to the emotion tag;

[0019] Clustering the personality information of the historical conversation texts belonging to the same interlocutor according to the identity identifier to obtain the personality description content of the interlocutor;

[0020] The personality description content is mapped to a preset personality space to determine the personality characteristics of different interlocutors in the historical dialogue text.

[0021] Optionally, the using the historical conversation text and the control text as positive and negative samples respectively to fine-tune the pre-trained language model includes:

[0022] Inputting the historical conversation text as a positive sample into the pre-trained language model to determine a first correctness prediction result of the interlocutor to whom the historical conversation text belongs having personality characteristics;

[0023] Inputting the control text as a negative sample into the pre-trained language model to determine a second correctness prediction result that the interlocutor to which the control text belongs does not have the personality characteristics;

[0024] According to the weight of the pre-trained language model under the current model parameters, the confidence probabilities of the positive sample and the negative sample are calculated respectively through the first correctness prediction result and the second correctness prediction result;

[0025] Taking the maximization of the confidence probability of the positive sample and the negative sample as the goal, fine-tuning the model parameters of the pre-trained language model is performed to obtain a target recognition model.

[0026] Optionally, obtaining the historical conversation text includes:

[0027] Get multiple sentence texts;

[0028] Determine the identity corresponding to the statement text according to the generation time of the statement text and the user address;

[0029] Inputting the sentence text into a sentiment recognition model to obtain a sentiment label corresponding to the text sentence;

[0030] According to the generation time of the sentence text, the sentence text and its corresponding identity identifier and emotion tag are combined in sequence to form a historical dialogue text carrying the emotion tag and identity identifier.

[0031] Optionally, the interaction method of the medical device further includes:

[0032] Performing natural language processing on the conversation sample to obtain word vectors of multiple word segments in the conversation sample;

[0033] Connecting the word vectors to form semantic features of the conversation sample;

[0034] Using a preset routing protocol, performing matrix transformation on the semantic features according to the activation vector, and determining the emotional features of the conversation sample;

[0035] The emotion recognition model is trained according to the emotion feature and the dialogue sequence relationship of the dialogue sample by using a loss function.

[0036] Optionally, the obtaining of feedback text and / or feedback action matching the voice text includes:

[0037] Matching the voice text with different preset interactive texts in a preset text library;

[0038] If there is a preset interactive text matching the voice text in the preset text library, filtering the feedback information from the preset interactive text according to the usage frequency and usage time;

[0039] If there is no preset interactive text matching the voice text in the preset text library, searching in a set network address according to the voice text, and using the filtered search result text as the feedback information;

[0040] The voice text is matched with different preset interactive actions in a preset action library, and the preset interactive actions matching the voice text are used as the feedback actions.

[0041] According to another aspect of the present application, there is provided an interactive device for a medical device, comprising:

[0042] A text conversion module, configured to convert the voice information into voice text in response to the medical device receiving the voice information;

[0043] A personality recognition module, used to input the voice text into a target recognition model to obtain the target personality characteristics of the outputter of the voice information, wherein the target recognition model is obtained by fine-tuning a pre-trained language model through the confidence probabilities of the interlocutors having and not having the personality characteristics in the historical conversation text;

[0044] A feedback module, used to obtain feedback text and / or feedback action matching the voice text;

[0045] An updating module, used for updating the vocabulary of the feedback text and / or the control parameters of the feedback action according to the target personality characteristics;

[0046] An execution module is used to output the updated feedback information and / or execute the feedback action.

[0047] Optionally, the interaction device of the medical device further includes:

[0048] An acquisition module, used to acquire the historical conversation text, wherein the historical conversation text carries an emotion tag and an identity identifier of the interlocutor;

[0049] The personality recognition module is further used to recognize personality characteristics of different interlocutors in the historical conversation text based on the emotion label and the identity identifier;

[0050] A construction module, used for constructing a comparison text without the individual characteristics based on the historical dialogue text;

[0051] The training module is used to use the historical conversation text and the control text as positive and negative samples respectively, to fine-tune the pre-trained language model, and obtain a target recognition model.

[0052] Optionally, the personality recognition module is specifically used to determine the personality information of the historical conversation text based on the emotional tag; cluster the personality information of the historical conversation text belonging to the same interlocutor based on the identity identifier to obtain the personality description content of the interlocutor; map the personality description content to a preset personality space to determine the personality characteristics of different interlocutors in the historical conversation text.

[0053] Optionally, the training module is specifically used to input the historical conversation text as a positive sample into the pre-trained language model to determine a first correctness prediction result that the interlocutor to which the historical conversation text belongs has personality characteristics; input the control text as a negative sample into the pre-trained language model to determine a second correctness prediction result that the interlocutor to which the control text belongs does not have personality characteristics; calculate the confidence probabilities of the positive sample and the negative sample respectively through the first correctness prediction result and the second correctness prediction result according to the weight of the pre-trained language model under the current model parameters; and fine-tune the model parameters of the pre-trained language model with the goal of maximizing the sum of the confidence probabilities of the positive sample and the negative sample to obtain a target recognition model.

[0054] Optionally, the acquisition module is specifically used to acquire multiple sentence texts; determine the identity identifier corresponding to the sentence text based on the generation time of the sentence text and the user address; input the sentence text into a sentiment recognition model to obtain the sentiment label corresponding to the text sentence; and combine the sentence text and its corresponding identity identifier and sentiment label in sequence based on the generation time of the sentence text to form a historical conversation text carrying the sentiment label and the identity identifier.

[0055] Optionally, the interaction device of the medical device further includes:

[0056] A semantic recognition module, used to perform natural language processing on the conversation sample to obtain word vectors of multiple word segments in the conversation sample; and to connect the word vectors to form semantic features of the conversation sample;

[0057] A vector transformation module, used to use a preset routing protocol to perform matrix transformation on the semantic features according to the activation vector to determine the emotional features of the conversation sample;

[0058] The training module is also used to train the emotion recognition model based on the emotion features and the dialogue sequence relationship of the dialogue samples by using a loss function.

[0059] Optionally, the feedback module is specifically used to match the voice text with different preset interactive texts in a preset text library; if there is a preset interactive text matching the voice text in the preset text library, the feedback information is filtered from the preset interactive text according to the frequency of use and the usage time; if there is no preset interactive text matching the voice text in the preset text library, a search is performed in a set network address according to the voice text, and the filtered search result text is used as the feedback information; the voice text is matched with different preset interactive actions in a preset action library, and the preset interactive action matching the voice text is used as the feedback action.

[0060] According to another aspect of the present application, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above-mentioned medical device interaction method are implemented.

[0061] According to another aspect of the present application, a medical device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, and the processor implements the steps of the above-mentioned interaction method of the medical device when executing the program.

[0062] Through the above technical solution, on the one hand, the historical dialogue text and speaker's character features are learned through the pre-trained language model, the speaker's language behavior in the dialogue is linked to the description of personality features, and the binary classification accuracy of the interlocutor with and without personality features is used to identify the personality of the interlocutor, which not only realizes automated personality prediction, but also can make up for the impact caused by the lack of text data sets and improve the personality prediction effect. On the other hand, medical devices can make corresponding emotional responses based on the unique personality features reflected in the user's voice information, increase the emotional exchange and communication between medical devices and users, further enhance the ability of medical devices to understand and manage dialogues, ensure the personalization and high efficiency of services, and help improve user satisfaction and loyalty.

[0063] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0065] Figure 1 A schematic diagram showing a flow chart of an interaction method of a medical device provided in an embodiment of the present application is shown;

[0066] Figure 2 A structural block diagram of an interactive device for medical equipment provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0067] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0068] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0069] Those skilled in the art will appreciate that, unless expressly stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, or there may be intermediate elements. In addition, the "connection" or "connection" used herein may include wireless connection or wireless fusion. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0070] Now, exemplary embodiments according to the present application will be described in more detail with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in a variety of different forms and should not be interpreted as being limited to the embodiments set forth herein. It should be understood that these embodiments are provided to make the disclosure of the present application thorough and complete, and to fully convey the concepts of these exemplary embodiments to those of ordinary skill in the art.

[0071] In this embodiment, a medical device interaction method is provided, such as Figure 1 As shown, the method includes:

[0072] Step 101, in response to a medical device receiving voice information, converting the voice information into voice text.

[0073] In this embodiment, when the medical device receives the voice from the user, the voice information is converted into voice text through a voice recognition algorithm, so that the medical device can quickly record the user's symptom description, medical history, consultation question description, event statement and other information, so as to better understand the user's needs and provide corresponding feedback, thereby improving the quality of subsequent treatment and nursing work.

[0074] Exemplarily, medical equipment includes consulting room interpretation robots, medical intelligent chat robots, medical testing equipment with human-computer dialogue functions, nursing robots, medical guidance robots, and other medical equipment that can interact with users.

[0075] Step 102, input the speech text into the target recognition model to obtain the target personality characteristics of the outputter of the speech information.

[0076] Among them, the target recognition model is obtained by fine-tuning the pre-trained language model through the confidence probability of the interlocutors having and not having personality characteristics in the historical dialogue text.

[0077] Specifically, the target personality traits include personality traits from the Big Five personality trait theory proposed by psychological research, namely, openness (O), conscientiousness (C), extraversion (E), agreeableness (A), and neuroticism (N). Taking the personality trait of agreeableness as an example, its evaluation criteria include affinity, cooperation, empathy, and compassion, and often give priority to harmonious relationships and the happiness of others; on the contrary, the evaluation criteria of not having agreeableness include aggressiveness, uncooperativeness, lack of empathy, and often putting one's own needs and desires above the happiness of others. Taking the personality trait of conscientiousness as an example, its evaluation criteria include being organized, responsible, hardworking, and detail-oriented, and being committed to achieving one's goals with a strong sense of responsibility and self-discipline; on the contrary, the evaluation criteria of not having responsibility include being disorganized, careless, impulsive, and lacking discipline, and often showing indifference to responsibility and commitment. For example, the characteristics of extroversion include being outgoing, sociable, energetic, thriving in social interactions, often seeking excitement and enjoying the company of others; on the contrary, the characteristics of non-extroversion include being introverted, reserved, quiet, often preferring to be alone or in a smaller social environment, conserving energy and finding satisfaction in self-reflection. For example, neuroticism includes being prone to negative emotions such as anxiety, worry, and mood swings, often showing high sensitivity to stress, and tending to be self-doubting and emotionally unstable; on the contrary, the characteristics of non-neuroticism include being emotionally stable, resilient, calm, often showing a calm and balanced demeanor, and tending to cope with stress and adversity easily. For example, the characteristics of openness include being curious, open-minded, creative, tolerant, emotionally expressive, and willing to accept new experiences and ideas; on the contrary, the characteristics of non-openness include being closed-minded, resisting change, lacking curiosity, intolerant of differences, emotionally reserved, and hesitant to explore new ideas or experiences.

[0078] In one embodiment, fine-tuning the pre-trained language model specifically includes the following steps:

[0079] Step 201, obtaining historical conversation text.

[0080] The historical conversation text carries emotion tags and interlocutor identities. The emotion tags are used to indicate the emotions carried by the interlocutor in each sentence in the historical conversation text, such as anger, disgust, fear, joy, sadness, surprise, neutrality, etc. The identity is used to distinguish different interlocutors in the conversation and is unique.

[0081] In an actual application scenario, step 201, i.e. obtaining the historical conversation text, specifically includes the following steps:

[0082] Step 201 - 1 , obtaining multiple sentence texts.

[0083] Step 201-2, determining the identity identifier corresponding to the sentence text according to the generation time of the sentence text and the user address.

[0084] Step 201 - 3 , input the sentence text into the emotion recognition model to obtain the emotion label corresponding to the text sentence.

[0085] Step 201-4, according to the generation time of the sentence text, the sentence text and its corresponding identity identifier and emotion tag are combined in sequence to form a historical dialogue text carrying the emotion tag and identity identifier.

[0086] In this embodiment, the source of each sentence is clearly determined and the different speakers of multiple sentence texts are distinguished by the time when the sentence text is generated and the user address. Each speaker is assigned a unique identity, which avoids confusion caused by unclear speakers and makes the subsequent analysis based on the content of the conversation more accurate. At the same time, the emotional state expressed by the speaker when each text sentence is generated is identified through the trained emotion recognition model, and the corresponding emotion label is marked for each text sentence, so as to better interpret the potential information behind the sentence, thereby more accurately analyzing the content of the conversation. Finally, based on the time when the sentence text was generated, multiple sentence texts and their corresponding identity labels and emotion labels are arranged and combined to form a historical conversation text carrying emotion labels and identity labels, which helps to expand the historical conversation text to the personality recognition task that supports multi-person conversation scenarios.

[0087] It is worth mentioning that the emotion recognition model can be trained by the following steps: perform natural language processing on the dialogue samples to obtain word vectors of multiple word segments in the dialogue samples; connect the word vectors to form the semantic features of the dialogue samples; use the preset routing protocol to perform matrix transformation on the semantic features according to the activation vector to determine the emotional features of the dialogue samples; use the loss function to train the emotion recognition model based on the emotional features and the dialogue sequence relationship of the dialogue samples.

[0088] The activation vector represents the instantiation parameters of a specific type of entity. The length of the activation vector is the possibility of the entity's existence, and the direction of the activation vector is the direction of the entity.

[0089] In this embodiment, the text semantics of the dialogue samples are extracted at the word level and spliced ​​together to obtain the semantic features of the dialogue samples. The semantic features are used as a group of low-level neurons, and the instantiation parameters of the high-level neurons are converted through activation vectors and transformation matrices to obtain the emotional features of the dialogue samples. The emotional features contain language information related to the task, while the dialogue sequence relationship of the dialogue sample represents the general grammatical logic. The two are combined to form the final feature vector to train the emotion recognition model. In this way, the complex emotional feature vector is embedded in the dialogue sequence relationship for joint encoding, so as to obtain an emotional understanding result that is more in line with the contextual information of the dialogue context and improve the accuracy of emotion recognition.

[0090] Exemplarily, the following algorithm is used to perform matrix transformation on semantic features according to the activation vector:

[0091]

[0092] Among them, v j represents the output of neuron j, s j Represents the total input of the neuron. Except for the first layer of neurons, the total input of each subsequent layer of neurons is the weighted sum of all prediction vectors of the previous layer of neurons. The prediction vector requires a weight matrix W ij Multiply the predicted vector ui of the actual output of the previous layer to get it. ij is the coupling coefficient determined during the dynamic routing process, all c ij The sum is 1. b ij The initial value of is the logarithmic prior probability, which is used to represent the possibility of coupling between neuron i and neuron j. The initial coupling coefficient can be determined by the similarity between vectors.

[0093] Step 202, identifying the personality characteristics of different interlocutors in the historical conversation text based on the emotion tags and identity identifiers.

[0094] In this embodiment, the conversation content is enhanced by discrete emotion tags and identity identifiers of historical conversation content, thereby effectively capturing the interactive relationship between emotional features and context, so that the pre-trained language model can learn the personality characteristics and language preferences of a specific speaker, thereby improving the recognition accuracy and interpretability of the interlocutor's personality characteristics.

[0095] In an actual application scenario, step 202, that is, identifying the personality characteristics of different interlocutors in the historical conversation text based on the emotion label and the identity identifier, specifically includes the following steps:

[0096] Step 202-1, determining the personality information of the historical dialogue text based on the emotion tag.

[0097] Step 202-2, clustering the personality information of the historical conversation texts belonging to the same interlocutor according to the identity identifier to obtain the personality description content of the interlocutor.

[0098] Step 202-3, mapping the personality description content to a preset personality space, and determining the personality characteristics of different interlocutors in the historical dialogue text.

[0099] In this embodiment, the emotion label of each sentence in the historical dialogue text is matched with different personality evaluation standards, and the personality information that the interlocutor of each sentence in the historical dialogue text may have is determined, that is, the personality evaluation standard hit by the emotion label. The multi-faceted personality clues belonging to the same interlocutor are integrated to form the personality description content of the interlocutor. The personality description content is mapped to the preset personality space to determine the personality characteristics of different interlocutors in the historical dialogue text. In this way, the inner psychological characteristics of the interlocutor can be deeply excavated from the emotional dimension, the semantic understanding of the personality itself by the historical dialogue text can be enhanced, clues can be provided for subsequent model fine-tuning, and the interpretability of the results can be reduced. In addition, the personality expression of the interlocutor at different stages and in different situations can be captured, reducing the dilution effect of the influence of long-distance historical dialogue on the interlocutor's emotions.

[0100] Step 203, constructing a comparison text without individual characteristics based on the historical dialogue text.

[0101] Step 204, using the historical conversation text and the control text as positive and negative samples respectively, fine-tuning the pre-trained language model to obtain a target recognition model.

[0102] In this embodiment, after determining the personality characteristics of different interlocutors in the historical dialogue text, the personality characteristics of any interlocutor are taken as the object, and the historical dialogue text is used as the basis to construct a control text corresponding to the historical dialogue text output by the interlocutor when the interlocutor does not have the personality characteristics through language simulation. The historical dialogue text and the control text are used as positive and negative samples respectively to fine-tune the pre-trained language model. Through the binary opposition classification task, not only can the pre-trained language model learn the various meaning expressions and language characteristics of users in real dialogue scenarios, but also the control text can be used to assist the model in identifying expressions that do not conform to the language style of personality characteristics. Even in the absence of a data set, the boundaries of personality recognition can be clarified, so that the pre-trained language model has a more comprehensive and accurate understanding of the user's personality, so that the medical service provider can more accurately grasp the unique personality of each user, and thus better adapt to the real dialogue scenario.

[0103] In actual application scenarios, step 204, that is, using the historical conversation text and the control text as positive and negative samples respectively, fine-tuning the pre-trained language model to obtain a target recognition model, specifically includes the following steps:

[0104] Step 204-1, input the historical conversation text as a positive sample into the pre-trained language model to determine the first correctness prediction result that the interlocutor to whom the historical conversation text belongs has the personality characteristics.

[0105] Step 204-2, input the control text as a negative sample into the pre-trained language model to determine a second correctness prediction result that the interlocutor to which the control text belongs does not have the personality characteristics.

[0106] Among them, the pre-trained language model can be a BERT (Bidirectional Encoder Representations from Transformers) model, an ALBERT (A Lite BERT) model, a RoBERTa (Robustly OptimizedBERT Approach) model, a T5 (Text-To-Text Transfer Transformer) model, etc., which are not listed one by one in the embodiments of the present application.

[0107] Step 204 - 3 , according to the weight of the pre-trained language model under the current model parameters, the confidence probabilities of the positive samples and the negative samples are calculated respectively through the first correctness prediction result and the second correctness prediction result.

[0108] Step 204 - 4 , with the goal of maximizing the sum of the confidence probabilities of positive samples and negative samples, fine-tune the model parameters of the pre-trained language model to obtain a target recognition model.

[0109] In this embodiment, after the historical conversation text and the control text are input as positive and negative samples to the pre-trained language model, the personality recognition results of the pre-trained language model for the historical conversation text and the control text can be obtained. Since the personality characteristics of the interlocutor to whom the historical conversation text belongs are known. By comparing the personality recognition results with the known personality characteristics of the interlocutor, the first correctness prediction results and the second correctness prediction results of whether the interlocutor to whom the historical conversation text belongs has or does not have the personality characteristics can be determined. The weight of the pre-trained language model under the current model parameters is used as a coefficient, and the confidence probability of the positive sample is calculated by the first correctness prediction result, and the confidence probability of the negative sample is calculated by the second correctness prediction result. With the goal of maximizing the sum of the confidence probabilities of the positive and negative samples, the model parameters of the pre-trained language model are fine-tuned and trained, and the pre-trained language model is fine-tuned in a fast adjustment manner. Thereby, the usability of the fine-tuned target recognition model is improved without sacrificing accuracy. The fine-tuned target recognition model can link the speaker's language behavior in the conversation with the description of the personality characteristics, and provide an interpretable recognition result.

[0110] Exemplarily, the historical conversation text X is packaged into the following two input samples (positive sample and negative sample), and the pre-trained language model M is allowed to perform a MASK on the correctness of the input sample, and fill in the prediction results "yes" and "no" in [MASK].

[0111] T pos (X,p)=X af, H pos (p); Is this correct? [MASK].

[0112] T neg (X,p)=X af ,H neg (p); Is this correct? [MASK].

[0113] Among them, T pos (X,p) and T neg (X, p) represent the personality recognition results of positive samples and negative samples respectively. af represents the historical conversation text with emotion tags, X af ={(u1,a1),(u2,a2),…,(u m , a m )},u m Represents the sentence text, a m Indicates the sentiment label. pos (p) and H neg (p) represents the characteristic evaluation criteria for having and not having a specific personality trait p, respectively.

[0114] {T pos (X,p),yes},{T neg (X,p),no}; if y p =1.

[0115] {T pos (X,p),no},{T neg (X,p),yes}; if y p =0.

[0116] Among them, y p Indicates personality traits, y p ∈[y a ,y c ,y e ,y o ,y n ].

[0117] According to the input sample, the confidence probability in four cases is expressed as follows:

[0118] P pos (yes)=P([MASK]=yes|Mθ (T pos (X,p))).

[0119] P pos (no) = P ([MASK] = no | M θ (T pos (X,p))).

[0120] P neg (yes)=P([MASK]=yes|M θ (T neg (X,p))).

[0121] P neg (no) = P ([MASK] = no | M θ (T neg (X,p))).

[0122] Among them, P pos (yes) and P pos (no) respectively represent the confidence probability that the model correctly or incorrectly identifies the interlocutor with personality characteristics for the positive sample. neg (yes) and P neg (no) respectively represent the confidence probability that the model correctly or incorrectly identifies the interlocutor as not having personality characteristics for negative samples. θ Represents the weights of the model parameters.

[0123] The training goal is to fine-tune the parameters θ in M ​​to maximize the sum of the two probabilities, expressed as:

[0124] (P pos (no)+P neg (no))×y p +(P neg (yes)+P neg (no))×(1-y p ).

[0125] In inference, the outcome is determined by comparing the sum of the following probabilities:

[0126] P neg (yes)+P pos (no), P pos (yes)+P neg (no).

[0127] Step 103: Obtain feedback text and / or feedback action matching the voice text.

[0128] In an actual application scenario, step 103, i.e., obtaining feedback text and / or feedback action matching the voice text, specifically includes the following steps:

[0129] Step 103 - 1 , matching the voice text with different preset interactive texts in a preset text library.

[0130] Step 103 - 2 : If there is a preset interactive text matching the voice text in the preset text library, the feedback information is filtered from the preset interactive text according to the usage frequency and usage time.

[0131] Step 103-3: If there is no preset interactive text matching the voice text in the preset text library, a search is performed in the set network address according to the voice text, and the filtered search result text is used as feedback information.

[0132] The set network address may be the website address of a medical institution, medical research institution, patient forum, etc., in order to obtain more authoritative medical-related information.

[0133] Step 103 - 4 , matching the voice text with different preset interactive actions in the preset action library, and taking the preset interactive actions that match the voice text as feedback actions.

[0134] In this embodiment, for text-type feedback information, feedback prediction is performed in advance on the text that the user may provide and a variety of preset interactive texts are stored. The voice text is matched with the preset interactive text. If there is a preset interactive text matching the voice text in the preset text library, one is extracted from the preset interactive text as feedback information based on the frequency of use and the time of use. In this way, the diversity of information feedback is achieved while ensuring that the feedback content can meet the needs of the user. Similarly, the preset interactive action matching the voice text is searched in the preset action library as the feedback action.

[0135] Furthermore, if suitable feedback information is found in the pre-configured preset text library, the voice text can be used as a question to perform a network search in a set network address to obtain content outside the preset text library, so as to provide the user with more comprehensive feedback.

[0136] Step 104: Update the vocabulary of the feedback text and / or the control parameters of the feedback action according to the target personality characteristics.

[0137] In this embodiment, after determining the feedback text and / or feedback action that responds to the user's needs, the vocabulary of the feedback text and / or the control parameters of the feedback action are modified accordingly based on the user's target personality characteristics, so that the medical device can explicitly express behaviors that are consistent with the target personality characteristics when communicating and interacting with the user, thereby increasing the emotional exchange and communication between the medical device and the user.

[0138] Step 105: output updated feedback information and / or execute feedback action.

[0139] Specifically, text-based feedback information can be output in the form of voice broadcast or displayed in the form of graphic text.

[0140] The interactive method of medical devices provided in this application, on the one hand, learns historical conversation texts and speaker's character features through pre-trained language models, links the speaker's language behavior in the conversation with the description of personality features, and uses the binary classification accuracy of the interlocutor with and without personality features to identify the personality of the interlocutor, which not only realizes automated personality prediction, but also can make up for the impact caused by the lack of text data sets and improve the personality prediction effect. On the other hand, the medical device can make corresponding emotional responses based on the unique personality features reflected in the user's voice information, increase the emotional exchange and communication between the medical device and the user, make up for the fact that the existing technology only focuses on the contextual information in the conversation and ignores the speaker's perception information reflecting the personality characteristics of the person, further enhance the ability of the medical device to understand and manage the conversation, ensure the personalization and high efficiency of medical services, and help improve the success rate of treatment or feedback.

[0141] For example, in a medical device dialogue scenario, if a user is identified as exhibiting neurotic personality traits during the communication process, the medical device can adjust the dialogue strategy, such as using more humorous language to communicate, or recommending psychological counseling-related medical resources to better adapt to the user's personality.

[0142] It should be noted that the size of the serial numbers of the steps in the above embodiments does not mean 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 application.

[0143] Furthermore, if Figure 2 As shown, as a specific implementation of the above-mentioned interaction method of the medical device, an embodiment of the present application provides an interaction device 400 of a medical device, and the interaction device 400 of the medical device includes: a text conversion module 401, a personality recognition module 402, a feedback module 403, an update module 404 and an execution module 405.

[0144] The text conversion module 401 is used to convert the voice information into voice text in response to the medical device receiving the voice information;

[0145] The personality recognition module 402 is used to input the speech text into the target recognition model to obtain the target personality characteristics of the outputter of the speech information, wherein the target recognition model is obtained by fine-tuning the pre-trained language model according to the confidence probability of the interlocutor having or not having the personality characteristics in the historical conversation text;

[0146] Feedback module 403, used to obtain feedback text and / or feedback action matching the speech text;

[0147] An updating module 404, for updating the vocabulary of the feedback text and / or the control parameters of the feedback action according to the target personality characteristics;

[0148] The execution module 405 is used to output updated feedback information and / or execute feedback actions.

[0149] Furthermore, the interactive device 400 of the medical device also includes: an acquisition module (not shown in the figure), which is used to acquire historical conversation texts, wherein the historical conversation texts carry emotional labels and identity identifiers of the interlocutors; a personality recognition module 402, which is also used to identify the personality characteristics of different interlocutors in the historical conversation texts based on the emotional labels and identity identifiers; a construction module (not shown in the figure), which is used to construct a control text without personality characteristics based on the historical conversation texts; and a training module (not shown in the figure), which is used to use the historical conversation texts and the control texts as positive and negative samples, respectively, to fine-tune the pre-trained language model to obtain a target recognition model.

[0150] Furthermore, the personality recognition module 402 is specifically used to determine the personality information of the historical dialogue text based on the emotion tag; cluster the personality information of the historical dialogue text belonging to the same interlocutor based on the identity identifier to obtain the personality description content of the interlocutor; map the personality description content to a preset personality space to determine the personality characteristics of different interlocutors in the historical dialogue text.

[0151] Furthermore, the training module is specifically used to input the historical conversation text as a positive sample into the pre-trained language model to determine a first correctness prediction result that the interlocutor to which the historical conversation text belongs has personality characteristics; input the control text as a negative sample into the pre-trained language model to determine a second correctness prediction result that the interlocutor to which the control text belongs does not have personality characteristics; according to the weight of the pre-trained language model under the current model parameters, the confidence probabilities of the positive samples and the negative samples are calculated respectively through the first correctness prediction result and the second correctness prediction result; with the goal of maximizing the sum of the confidence probabilities of the positive samples and the negative samples, the model parameters of the pre-trained language model are fine-tuned to obtain a target recognition model.

[0152] Furthermore, the acquisition module is specifically used to acquire multiple sentence texts; determine the identity identifier corresponding to the sentence text based on the generation time of the sentence text and the user address; input the sentence text into the emotion recognition model to obtain the emotion label corresponding to the text sentence; and combine the sentence text and its corresponding identity identifier and emotion label in sequence based on the generation time of the sentence text to form a historical conversation text carrying the emotion label and identity identifier.

[0153] Furthermore, the interactive device 400 of the medical device also includes: a semantic recognition module (not shown in the figure), which is used to perform natural language processing on the dialogue sample to obtain word vectors of multiple word segments in the dialogue sample; and to connect the word vectors to form semantic features of the dialogue sample; a vector transformation module (not shown in the figure), which is used to adopt a preset routing protocol to perform matrix transformation on the semantic features according to the activation vector to determine the emotional features of the dialogue sample; and a training module, which is also used to use a loss function to train the emotion recognition model based on the emotional features and the dialogue sequence relationship of the dialogue sample.

[0154] Furthermore, the feedback module 403 is specifically used to match the voice text with different preset interactive texts in the preset text library; if there is a preset interactive text matching the voice text in the preset text library, feedback information is filtered from the preset interactive text based on the frequency of use and the usage time; if there is no preset interactive text matching the voice text in the preset text library, a search is performed in the set network address based on the voice text, and the filtered search result text is used as feedback information; the voice text is matched with different preset interactive actions in the preset action library, and the preset interactive action matching the voice text is used as the feedback action.

[0155] For the specific definition of the interaction device of the medical device, please refer to the definition of the interaction method of the medical device above, which will not be repeated here. Each module in the above-mentioned interaction device of the medical device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the medical device in the form of hardware, or can be stored in the memory of the medical device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0156] Based on the above Figure 1 The method shown in the embodiment of the present application accordingly provides a readable storage medium on which a computer program is stored, and when the program is executed by a processor, the above-mentioned Figure 1 Interaction method of medical device shown.

[0157] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.), including a number of instructions for enabling a medical device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0158] Based on the above Figure 1 The method shown, and Figure 2In order to achieve the above-mentioned purpose, the embodiment of the present application further provides a medical device, the computer device comprising a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-mentioned Figure 1 Interaction method of medical device shown.

[0159] Optionally, the medical device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), etc.

[0160] Those skilled in the art will appreciate that the medical device structure provided in this embodiment does not constitute a limitation on the medical device, and may include more or fewer components, or a combination of certain components, or different component arrangements.

[0161] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages and saves the hardware and software resources of the computer device, and supports the operation of information processing programs and other software and / or programs. The network communication module is used to realize communication between the components inside the storage medium, and communication with other hardware and software in the physical device.

[0162] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by means of software plus necessary general hardware platforms, or by hardware, in response to the medical device receiving voice information, converting the voice information into voice text; inputting the voice text into the target recognition model to obtain the target personality characteristics possessed by the outputter of the voice information, wherein the target recognition model is obtained by fine-tuning the pre-trained language model through the confidence probability of the interlocutor having and not having the personality characteristics in the historical conversation text; obtaining feedback text and / or feedback actions matching the voice text; updating the vocabulary of the feedback text and / or the control parameters of the feedback action according to the target personality characteristics; outputting the updated feedback information and / or executing the feedback action. In the embodiment of the present application, on the one hand, the historical conversation text and the speaker's character characteristics are learned through the pre-trained language model, the speaker's language behavior in the conversation is linked to the description of the personality characteristics, and the binary classification accuracy of the interlocutor having and not having the personality characteristics is used to perform the personality recognition of the conversation character, which not only realizes the automatic personality prediction, but also can make up for the impact caused by the lack of text data sets, and improve the personality prediction effect. On the other hand, medical devices can make corresponding emotional responses based on the unique personality characteristics reflected in the user's voice information, which increases the emotional exchange and communication between medical devices and users, further enhances the ability of medical devices to understand and manage conversations, ensures personalized and efficient services, and helps improve user satisfaction and loyalty.

[0163] Those skilled in the art will appreciate that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily necessary for implementing the present application. Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the description of the implementation scenario, or can be changed accordingly and located in one or more devices different from the present implementation scenario. The modules of the above-mentioned implementation scenario can be combined into one module, or can be further split into multiple submodules.

[0164] The above serial numbers of this application are only for description and do not represent the advantages and disadvantages of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of this application, but this application is not limited to them, and any changes that can be thought of by technicians in this field should fall within the scope of protection of this application.

Claims

1. A medical device interaction method, characterized in that: The method comprises: In response to the medical device receiving the voice information, converting the voice information into voice text; Inputting the speech text into a target recognition model to obtain the target personality characteristics of the outputter of the speech information, wherein the target recognition model is obtained by fine-tuning a pre-trained language model through the confidence probabilities of the interlocutor having and not having the personality characteristics in the historical conversation text; Acquire feedback text and / or feedback action matching the voice text; updating the vocabulary of the feedback text and / or the control parameters of the feedback action according to the target personality characteristics; Output the updated feedback information and / or execute the feedback action.

2. The medical device interaction method according to claim 1, characterized in that: The method further comprises: Acquire the historical conversation text, wherein the historical conversation text carries an emotion tag and an identity identifier of the interlocutor; Identifying personality characteristics of different interlocutors in the historical conversation text based on the emotion tag and the identity identifier; Constructing a control text without the individual characteristics based on the historical dialogue text; The historical conversation text and the control text are used as positive and negative samples respectively, and the pre-trained language model is fine-tuned to obtain a target recognition model.

3. The medical device interaction method according to claim 2, characterized in that: The identifying the personality characteristics of different interlocutors in the historical conversation text based on the emotion label and the identity identifier includes: Determining personality information of the historical dialogue text according to the emotion tag; Clustering the personality information of the historical conversation texts belonging to the same interlocutor according to the identity identifier to obtain the personality description content of the interlocutor; The personality description content is mapped to a preset personality space to determine the personality characteristics of different interlocutors in the historical dialogue text.

4. The medical device interaction method according to claim 2, characterized in that: The method of using the historical conversation text and the control text as positive and negative samples respectively to fine-tune the pre-trained language model includes: Inputting the historical conversation text as a positive sample into the pre-trained language model to determine a first correctness prediction result of the interlocutor to whom the historical conversation text belongs having personality characteristics; Inputting the control text as a negative sample into the pre-trained language model to determine a second correctness prediction result that the interlocutor to which the control text belongs does not have the personality characteristics; According to the weight of the pre-trained language model under the current model parameters, the confidence probabilities of the positive sample and the negative sample are calculated respectively through the first correctness prediction result and the second correctness prediction result; With the goal of maximizing the sum of the confidence probabilities of the positive samples and the negative samples, fine-tuning training is performed on the model parameters of the pre-trained language model to obtain the target recognition model.

5. The medical device interaction method according to any one of claims 1 to 4, characterized in that: The method further comprises: Get multiple sentence texts; Determine the identity corresponding to the statement text according to the generation time of the statement text and the user address; Inputting the sentence text into a sentiment recognition model to obtain a sentiment label corresponding to the text sentence; According to the generation time of the sentence text, the sentence text and its corresponding identity identifier and emotion tag are combined in sequence to form the historical conversation text carrying the emotion tag and identity identifier.

6. The medical device interaction method according to claim 5, characterized in that: The method further comprises: Performing natural language processing on the conversation sample to obtain word vectors of multiple word segments in the conversation sample; Connecting the word vectors to form semantic features of the conversation sample; Using a preset routing protocol, performing matrix transformation on the semantic features according to the activation vector, and determining the emotional features of the conversation sample; The emotion recognition model is trained according to the emotion feature and the dialogue sequence relationship of the dialogue sample by using a loss function.

7. The medical device interaction method according to any one of claims 1 to 4, characterized in that: The obtaining of feedback text and / or feedback action matching the speech text includes: Matching the voice text with different preset interactive texts in a preset text library; If there is a preset interactive text matching the voice text in the preset text library, filtering the feedback information from the preset interactive text according to the usage frequency and usage time; If there is no preset interactive text matching the voice text in the preset text library, searching in a set network address according to the voice text, and using the filtered search result text as the feedback information; The voice text is matched with different preset interactive actions in a preset action library, and the preset interactive actions matching the voice text are used as the feedback actions.

8. An interactive device for medical equipment, characterized in that: The device comprises: A text conversion module, configured to convert the voice information into voice text in response to the medical device receiving the voice information; A personality recognition module, used to input the voice text into a target recognition model to obtain the target personality characteristics of the outputter of the voice information, wherein the target recognition model is obtained by fine-tuning a pre-trained language model through the confidence probabilities of the interlocutors having and not having the personality characteristics in the historical conversation text; A feedback module, used to obtain feedback text and / or feedback action matching the voice text; An updating module, used for updating the vocabulary of the feedback text and / or the control parameters of the feedback action according to the target personality characteristics; An execution module is used to output the updated feedback information and / or execute the feedback action.

9. A readable storage medium having a program or instruction stored thereon, characterized in that: When the program or instruction is executed by a processor, the steps of the medical device interaction method according to any one of claims 1 to 7 are implemented.

10. A medical device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that: When the processor executes the program, the interaction method of the medical device according to any one of claims 1 to 7 is implemented.

Images