A method for correcting elderly people's conversation intention information based on deep neural network

Abstract

The present invention discloses a method for correcting conversation intention information of the elderly based on a deep neural network, which specifically relates to the technical field of conversation optimization. The method comprises the following steps: collecting and cleaning multi-round conversation data on multiple topics, annotating the multi-round conversation corpus and unifying the format; performing feature extraction on the annotated corpus using a multi-intention recognition sub-model to generate a multi-intention representation vector capable of representing each intention; screening contextual content related to the target intention by combining a hierarchical attention mechanism with dynamic weight adjustment of user context; processing the contextual content related to the target intention through a gating operation based on multi-round interaction history, and retaining semantic fragments with reference value; performing parallel analysis on the reference semantic fragments based on a multi-task learning structure of a deep neural network, classifying and processing different intentions respectively, and calculating scores; performing a multi-intention fusion correction operation based on the classification and scoring results, and outputting intent information capable of distinguishing each intention.

Description

Technical Field

[0001] The present invention relates to the technical field of dialogue optimization, and more specifically, to a method for correcting conversation intention information of elderly people based on a deep neural network. Background Art

[0002] Current dialogue systems primarily rely on keyword matching or simple intent recognition algorithms to understand user intent. While these systems perform well for everyday conversations, they have significant limitations when used with elderly users. Elderly conversations often involve multiple, complex, and context-specific intents, such as health consultations, family arrangements, or lifestyle advice, which can all occur simultaneously within a single conversation. Furthermore, elderly users' language often contains ambiguous, implicit, or incomplete information. Coupled with their strong focus on health and family background, traditional dialogue systems struggle to meet their actual needs.

[0003] In the existing technology, it is difficult to effectively distinguish and accurately correct the information of each intent in a conversation where multiple intents are expressed in parallel, which easily leads to intention confusion and deviation, and thus cannot meet the needs of elderly users. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method for correcting the conversation intention information of the elderly based on a deep neural network to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] The method for correcting the elderly's conversation intention information based on a deep neural network includes the following steps:

[0007] S1: Collect and clean multi-round conversation data containing multiple topics, annotate the multi-round conversation data and unify the format;

[0008] S2: Use the multi-intent recognition sub-model to extract topic features from the annotated multi-round dialogue data and generate multi-intent representation vectors that can represent each intent;

[0009] S3: A hierarchical attention mechanism with dynamic weight adjustment based on user context is used to perform correlation calculations on multiple intent representation vectors to filter contextual content related to the target intent.

[0010] S4: Process the contextual content related to the target intent through gating operations based on multi-round interaction history, and retain reference semantic fragments that are valuable for intent correction;

[0011] S5: A multi-task learning structure based on a deep neural network analyzes the reference semantic fragments in parallel, classifies different intents and calculates the score for each intent;

[0012] S6: Perform multi-intent fusion correction operations based on the results of the classification process and intent scoring, and output intent information that can distinguish each intent.

[0013] In a preferred embodiment, collecting and cleaning multi-round conversation data containing multiple topics, annotating the multi-round conversation data and unifying the format, specifically includes:

[0014] Obtain multi-turn conversation data from multi-channel data sources;

[0015] Clean the multi-round conversation data to remove noise, incomplete data, and meaningless conversation records.

[0016] Use predefined annotation rules to annotate the cleaned multi-turn dialogue data. The annotation rules include annotation of topic categories, intent categories, and dialogue turn information.

[0017] The labeled multi-round dialogue data is formatted in a unified manner. The format unification includes the normalization of data fields, the unified representation of time series, and the normalization of the order of dialogue turns.

[0018] In a preferred embodiment, the multi-intent recognition sub-model is used to extract topic features from the annotated multi-round dialogue data to generate a multi-intent representation vector that can represent each intent, specifically including:

[0019] By inputting the pre-trained multi-intent recognition sub-model into the annotated multi-round dialogue data, the text content of each round of dialogue is analyzed and semantic information related to the topic features is extracted;

[0020] In the multi-intent recognition sub-model, the semantic analysis module is used to extract features from the text content of each round of dialogue. The extracted features include semantic features of the topic category and semantic features of the intent category.

[0021] Use the embedding vector generation module to embed the extracted semantic features and generate initial vectors that can represent the topic category and intent category;

[0022] The feature fusion module of the multi-intent recognition sub-model is used to fuse the initial vectors generated in each round of dialogue to obtain a multi-intent representation vector that can comprehensively represent multiple intent categories and topic categories.

[0023] In a preferred embodiment, a hierarchical attention mechanism combined with dynamic weight adjustment of user context is used to perform association calculation on multiple intent representation vectors to filter contextual content related to the target intent, specifically including:

[0024] The multi-intent representation vector is layered according to the predefined attention level, and each layer assigns initial weights to different dimensional features in the user context information;

[0025] Dynamically adjust the weight distribution of attention at each layer based on the historical conversation content and semantic features of the current user context information;

[0026] In each layer of the attention mechanism, the multi-intent representation vectors are weighted according to the adjusted weight distribution to obtain the weight distribution of different intent categories in each layer;

[0027] Through the weighted results of the multi-layer attention mechanism and the contextual information related to the target intent, the semantic fragments with the strongest relevance to the current target intent are screened out;

[0028] The semantic fragments that are most relevant to the current target intent are processed uniformly to generate a contextual content representation that can characterize the target intent.

[0029] In a preferred embodiment, the specific formula for dynamically adjusting the weight distribution of attention in each layer is as follows: ;in, Indicates the adjusted layer attention weights, Indicates the The initial weights of the layers, Indicates user context information and The similarity of the multi-intention representation vector input of the layer, Represents user context information, Indicates the The multi-intent representation vector input of the layer.

[0030] In a preferred embodiment, the contextual content related to the target intent is processed by a gating operation based on multiple rounds of interaction history, and reference semantic segments with reference value for intent correction are retained, specifically including:

[0031] Extracting historical information about user interactions from multiple rounds of conversation from contextual content related to the target intent, including conversation rounds, contextual semantic fragments, and time series features associated with the target intent.

[0032] Perform feature screening on the extracted historical information of multiple rounds of user interactions, assign an initial weight to each piece of historical information based on its relevance to the target intent, and remove historical information whose relevance to the target intent is lower than a preset threshold;

[0033] Based on the initial weights, the contextual content related to the target intent is filtered through gating operations. Specifically, the gating threshold is dynamically adjusted according to the weight of each historical information to filter out the contextual content that meets the gating conditions.

[0034] The context content filtered out by the gating operation is normalized, including time series sorting and content structuring of the filtered context semantic fragments, and the normalized context content is stored as a reference semantic fragment.

[0035] In a preferred embodiment, a multi-task learning structure based on a deep neural network performs parallel analysis on the reference semantic segments, classifies different intents and calculates the score of each intent, specifically including:

[0036] Preprocess the reference semantic fragments according to their contextual relevance and time series characteristics;

[0037] Build a multi-task learning structure for deep neural networks, which includes intent classification and intent scoring. The intent classification task is used to identify the intent category corresponding to the reference semantic segment, and the intent scoring task is used to calculate the weight score for each intent.

[0038] In the multi-task learning structure, the pre-processed reference semantic segments are processed in parallel, and the intent classification task classifies each semantic segment to determine the intent category to which it belongs;

[0039] In the intent scoring task, each intent category is scored based on the semantic features and contextual relationships of the reference semantic fragment. The scoring is based on the degree of association between the semantic fragment and the target intent and the time series characteristics.

[0040] The results of intent classification and intent scoring are integrated to generate the classification label and scoring value for each reference semantic segment.

[0041] In a preferred embodiment, a multi-intent fusion correction operation is performed based on the results of the classification process and the intent scoring, and intent information that can distinguish each intent is output, specifically including:

[0042] Receive the classification labels and scoring values generated by the classification processing and intent scoring results, aggregate the classification labels according to the intent categories, and initially form a classification set of intent categories;

[0043] Based on the intent categories and their ratings in the classification set, sort the intent categories in descending order of ratings, and select the intent category with the highest rating as the primary intent.

[0044] Perform semantic supplementation on the intent categories with lower scores but certain relevance, and merge them with the main intent to form a comprehensive correction result of multiple intents;

[0045] The semantic information of the main intent and related intents is fused through a multi-intent fusion algorithm to generate intent information that can distinguish between the intents.

[0046] The fused and corrected intent information is structured and stored, and a clear classification result that can distinguish between the main intent and related intent is output.

[0047] The technical effects and advantages of the method of correcting the elderly's conversation intention information based on deep neural network in this invention are as follows:

[0048] 1. The present invention can effectively solve the intention confusion and deviation problems caused by the parallel expression of multiple intentions in conversations among the elderly through multi-task learning and hierarchical attention mechanism based on deep neural networks; by collecting and cleaning multi-round conversation data of the elderly on multiple topics, and using a multi-intention recognition sub-model to extract features from the labeled corpus, a multi-intention representation vector that can represent each intention is generated, laying the foundation for subsequent intent classification and correction. The hierarchical attention mechanism combined with dynamic weight adjustment of user context can accurately screen out contextual content related to the target intent, avoid interference of irrelevant information on the intent correction process, and thus improve the system's adaptability to complex contexts.

[0049] 2. Through gating operations based on multi-round interaction history, the contextual content related to the target intent is processed, and semantic fragments with reference value for intent correction are retained. At the same time, the multi-task learning structure of the deep neural network is used to perform parallel analysis on the reference semantic fragments, classify and process different intents, and calculate the scoring results, thereby realizing the dynamic fusion and correction operations of multiple intents; by outputting clear and structured intent information, it can effectively distinguish between various intents, improve the system's processing accuracy for the parallel expression of multiple intents, significantly optimize the interactive experience of elderly people's conversations, and provide reliable support for intelligent decision-making in scenarios such as health consultation and family arrangements. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 Schematic diagram of the method for correcting the conversation intention information of the elderly based on a deep neural network in the present invention. DETAILED DESCRIPTION

[0051] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0052] Example: Figure 1 The present invention provides a method for correcting elderly people's conversation intention information based on a deep neural network, which includes the following steps:

[0053] S1: Collect and clean multi-round dialogue data containing multiple topics, annotate the multi-round dialogue data and unify the format.

[0054] S2: Use the multi-intent recognition sub-model to extract topic features from the annotated multi-round dialogue data and generate a multi-intent representation vector that can represent each intent.

[0055] S3: A hierarchical attention mechanism combined with dynamic weight adjustment of user context is used to perform association calculations on multiple intent representation vectors to filter contextual content related to the target intent.

[0056] S4: Process the contextual content related to the target intent through gating operations based on multi-round interaction history, and retain reference semantic fragments that are of reference value for intent correction.

[0057] S5: A multi-task learning structure based on a deep neural network performs parallel analysis on the reference semantic fragments, classifies different intents separately, and calculates the score of each intent.

[0058] S6: Perform multi-intent fusion correction operations based on the results of the classification process and intent scoring, and output intent information that can distinguish each intent.

[0059] Collect and clean multi-round conversation data covering multiple topics, annotate the multi-round conversation data and unify the format, including:

[0060] Obtain multi-turn conversation data from multi-channel data sources:

[0061] Multi-turn conversation data is collected from multiple data sources, including social media platforms, health consultation systems, home voice assistant devices, and elderly care systems. Data sources should cover key topics commonly discussed by seniors, such as health, family, and daily life, to ensure data diversity and coverage. The collected data includes text conversations, speech-to-text transcripts, and corresponding conversation metadata, such as conversation duration, participant roles, and conversation context.

[0062] Clean the multi-round conversation data to remove noise, incomplete data, and meaningless conversation records:

[0063] Remove noisy data: Use rules or machine learning models to filter out text that is semantically incomplete, contains numerous spelling errors, or has no practical meaning, such as garbled characters or meaningless records consisting of single letters.

[0064] Eliminate incomplete data: Delete conversation records that are missing key information (such as conversation turn number, conversation content, timestamp, etc.) or are insufficient in length;

[0065] Delete meaningless conversation records: Identify and remove common sentence patterns without actual meaning, such as irrelevant conversations consisting of phrases such as "um" and "okay".

[0066] The cleaned multi-turn dialogue data is annotated using predefined annotation rules. The annotation rules include annotation of topic categories, intent categories, and dialogue turn information:

[0067] Labeling rules need to be defined in advance and developed by domain experts. Labeling includes the following:

[0068] Topic category labeling: label topics such as health, family, and daily life based on the conversation content;

[0069] Intent category annotation: Classify the explicit or implicit intent in each round of conversation, such as asking about health conditions or discussing family activities;

[0070] Dialogue turn information annotation: Add a turn number to each dialogue round to ensure that subsequent steps can accurately analyze the timing relationship of the dialogue.

[0071] The annotation work can be completed by combining manual annotation and automatic annotation. The automatic annotation part relies on semantic analysis algorithms to perform preliminary classification of common intentions, while manual annotation performs fine-grained processing on special or complex corpus.

[0072] The annotated multi-turn dialogue data is formatted in a unified manner. This includes normalizing data fields, unifying the time series representation, and normalizing the order of dialogue turns.

[0073] Normalize data fields: Ensure that the fields of each record are complete and structured, such as the unified definition and naming of the conversation turn field, topic category field, and intent category field;

[0074] Unified representation of time series: Standardize the timestamp format in all conversation records, for example, convert the "year-month-day-hour-minute" format to ISO standard time representation;

[0075] Normalize the sorting of conversation turns: Sort multi-turn conversation records according to the chronological order of the conversation to ensure that the data can accurately reflect the natural order and contextual relationships of the conversation.

[0076] The multi-intent recognition sub-model is used to extract topic features from the annotated multi-round dialogue data and generate multi-intent representation vectors that can represent each intent. Specifically, it includes:

[0077] By inputting the pre-trained multi-intent recognition sub-model into the annotated multi-round dialogue data, the text content of each round of dialogue is analyzed and semantic information related to the topic features is extracted:

[0078] The labeled multi-turn dialogue corpus is fed into the pre-trained multi-intent recognition sub-model as input data. The multi-turn dialogue corpus contains text content with clearly labeled topic categories, intent categories, and turn information.

[0079] The multi-intent recognition sub-model uses a semantic parsing algorithm to perform word segmentation, grammatical analysis, and semantic association processing on the text, ensuring accurate extraction of implicit semantic information within each round of dialogue. This text content parsing step includes: parsing each round of dialogue in the corpus separately to preserve the semantic integrity of the turn information; categorizing the parsed results by topic and intent categories to support subsequent module processing; and, for multi-round context alignment, recording the semantic connections between turns to ensure the relevance of multi-intent semantics.

[0080] In the multi-intent recognition sub-model, the semantic analysis module is used to extract features from the text content of each round of dialogue. The extracted features include semantic features of the topic category and semantic features of the intent category:

[0081] The topic classifier in the semantic analysis module categorizes the text content of each conversation into topics based on a predefined classification system (such as health, family, and daily life), and extracts feature information for each topic category. This extraction process, based on the training results of the word embedding model, records the semantic features of keywords or phrases related to the topic category.

[0082] After topic features are extracted, the intent classifier is used to categorize the text intent of each conversation. The intent classification system includes explicit intent (such as direct questions) and implicit intent (such as expressing emotions and asking rhetorical questions). Semantic features are extracted based on a deep semantic matching algorithm, which calculates the relevance weights between text segments and each intent category to extract important semantic features.

[0083] The extracted semantic features include the semantic representation of word vectors, context-related time series information, and the correspondence between topic categories and intent categories.

[0084] Use the embedding vector generation module to embed the extracted semantic features and generate initial vectors that can represent the topic category and intent category:

[0085] An embedding model is used to process the semantic features of topic categories to generate embedding vectors for the topic categories. Based on a pretrained word embedding algorithm, the embedding model generates fixed-length, high-dimensional vectors based on the input semantic features. The same embedding process is performed on the semantic features of intent categories to generate embedding vectors for the intent categories. This process takes into account the temporal sequence information of the conversation context to maintain the coherence of the semantic representation.

[0086] The embedding vector formula is as follows: ;in, The comprehensive embedding vector is a fusion representation obtained by weighted calculation of the semantic feature vectors of the topic category and the intent category, which is used for subsequent multi-intent analysis; It is the weight matrix of the semantic features of the topic category, which indicates the weight ratio of the topic category features when calculating the comprehensive embedding vector. It is dynamically adjusted by the multi-intent recognition sub-model during the training process. It is a vector representation of the semantic features of the topic category. It represents the high-dimensional vector obtained after the semantic features corresponding to the topic category are processed by the embedding vector generation module, which is used to reflect the topic information involved in the conversation; is the weight matrix of the semantic features of the intent category, which represents the importance weight of the intent category features in the calculation of the comprehensive embedding vector. The model will dynamically adjust the value of this matrix based on the training data; It is a vector representation of the semantic features of the intent category. It is converted from the semantic features of the intent category into a high-dimensional representation through the embedding vector generation module, and is used to characterize the semantic features of the intent category in the conversation.

[0087] The weight matrix in the embedding vector formula is dynamically updated by the semantic matching model through gradient descent during the training process to ensure that the weights can reflect the importance of different semantic features.

[0088] The feature fusion module of the multi-intent recognition sub-model fuses the initial vectors generated in each round of dialogue to obtain a multi-intent representation vector that can comprehensively represent multiple intent categories and topic categories:

[0089] Use the weighted formula to perform weighted summation on the embedding vectors of the topic category and intent category: ;in, Represents the final generated multi-intent representation vector; and are the fusion weight parameters for topic category and intent category, which are automatically adjusted by the feature fusion module; It is the global semantic feature vector of the multi-round dialogue context, calculated by the semantic alignment module.

[0090] The multi-intention representation vector after fusion calculation is normalized to ensure that the distribution range of the final vector in the high-dimensional space meets the input requirements of subsequent processing.

[0091] Notably, the multi-intent recognition sub-model consists of a semantic analysis module, an embedding vector generation module, and a feature fusion module, each with clear division of labor and coordinated operation. The semantic analysis module parses the input annotated multi-turn conversation corpus. Specifically, it includes a word segmentation unit, a grammatical parsing unit, and a semantic matching unit, sequentially performing word segmentation, grammatical structure analysis, and semantic feature extraction on the text content. The embedding vector generation module vectorizes the features extracted by the semantic analysis module. Specifically, it includes a word embedding sub-unit, a semantic embedding sub-unit, and a feature combination sub-unit. It generates high-dimensional vector representations of topic categories and intent categories using a pre-trained model. The feature fusion module further fuses the embedding vectors. Specifically, it includes a weighted fusion sub-unit, a normalization sub-unit, and a context alignment sub-unit. It weights the embedding vectors of topic categories and intent categories using predefined weights and semantically aligns them with the contextual information from the multi-turn conversation. Ultimately, it generates a comprehensive representation vector that represents multiple intent categories and multiple topic categories. Through the coordinated collaboration between these modules, the multi-intent recognition sub-model gradually transforms the annotated multi-turn conversation corpus into a multi-intent representation vector that can be used in subsequent steps.

[0092] A hierarchical attention mechanism with dynamic weight adjustment of user context is used to perform correlation calculations on multiple intent representation vectors to filter contextual content related to the target intent, including:

[0093] The multi-intent representation vectors are layered according to the predefined attention levels. Each layer assigns initial weights to different dimensional features in the user context information:

[0094] The initial weights are allocated based on the dimensional features of the contextual information. Each dimensional feature includes topic category information, intent category information, and time series features related to the conversation context. During the hierarchical processing, the attention level of each layer is defined as an independent computing unit. Each layer receives the input of multiple intent representation vectors and performs feature screening for specific dimensional features.

[0095] Through this hierarchical processing, the multi-intent representation vector is decomposed into multiple sub-vectors with specific dimensional features, which serve as inputs for subsequent weight adjustment and weighted calculation.

[0096] Dynamically adjust the weight distribution of attention at each layer based on the historical conversation content and semantic features of the current user context information:

[0097] First, semantic features are extracted from the user's historical conversation content. These features include topic continuity, contextual semantic relevance, and the potential contextual relationship of the target intent. The initial weights of each layer are dynamically adjusted based on the matching degree between the semantic features in the current user context information and the multi-intent representation vector. The specific formula for dynamic adjustment is as follows: ;in, Indicates the adjusted layer attention weights, Indicates the The initial weights of the layers, Indicates user context information and The similarity of the multi-intention representation vector input of the layer, Represents user context information, Indicates the The multi-intent representation vector input of the layer.

[0098] in, The calculation of is based on the cosine distance or dot product operation between the user semantic feature and the specific dimension of the representation vector, and the result is used as a dynamically adjusted scaling factor.

[0099] In each layer of the attention mechanism, the multi-intent representation vectors are weighted according to the adjusted weight distribution to obtain the weight distribution of different intent categories in each layer:

[0100] Use the adjusted attention weights to weight the multi-intent representation vectors of each layer input. The calculation formula is as follows: ;in, Indicates the Feature output after layer weighted calculation;

[0101] The weighted result retains the contextual content related to the target intention in each layer, and achieves hierarchical accumulation of semantic information through layer-by-layer weighting.

[0102] By combining the weighted results of the multi-layer attention mechanism and the contextual information related to the target intent, the semantic fragments with the strongest relevance to the current target intent are selected:

[0103] The weight distribution of the weighted results at each level is integrated, and the context information with the highest weight value is used as the candidate semantic segment with the strongest correlation with the target intent. The final matching value between the target intent and the candidate semantic segment is calculated. The screening formula is as follows: ;in, Represents the semantic fragments finally filtered out, Represent the output feature values of the multi-layer attention weighted results, Indicates the number of attention mechanism layers.

[0104] Through the above filtering operations, it is ensured that the semantic relevance between the filtered semantic fragments and the current target intention is the strongest.

[0105] The semantic fragments that are most relevant to the current target intent are processed uniformly to generate a contextual content representation that can represent the target intent:

[0106] The filtered semantic fragments are re-encoded to retain the semantic context; the encoding results are stored as a structured representation for input in subsequent steps.

[0107] Through gating operations based on multi-round interaction history, contextual content related to the target intent is processed, retaining reference semantic fragments that are valuable for intent correction, including:

[0108] Extract user multi-round interaction history information from the context content related to the target intent, including conversation turns, context semantic fragments, and time series features associated with the target intent:

[0109] Conversation turn information: records the sequence number of each user's conversation turn to ensure that subsequent analysis can be processed according to the natural context;

[0110] Contextual semantic fragments: Extract semantic fragments that are directly or indirectly related to the target intent, such as extracting "medication time" and "physical symptoms" in health consultation scenarios;

[0111] Time series features: record the time information of each round of dialogue, which is used for subsequent analysis of the temporal relationship of semantic segments.

[0112] The extracted multi-round interaction history information is organized in a structured format to ensure that each piece of information contains a complete round number, semantic fragment, and timestamp, providing basic data for subsequent feature screening and gating operations.

[0113] Perform feature screening on the extracted historical information of multiple rounds of user interactions, assign an initial weight to each piece of historical information based on its relevance to the target intent, and remove historical information whose relevance to the target intent is lower than a preset threshold:

[0114] Based on the characteristics of the target intent, the semantic relevance between historical information from multiple rounds of interactions and the target intent is analyzed. For example, in the "family arrangements" scenario, if the target intent is "determine the activity time," historical information containing keywords such as time and activity name is prioritized.

[0115] An initial weight is assigned to each piece of historical information. The weight is assigned based on the degree of semantic matching. For example, historical information containing target intent keywords or semantic features is assigned a higher weight; content whose semantic fragments have no obvious relevance to the target intent is assigned a lower weight.

[0116] Historical information with a weight below a preset threshold is removed, retaining only contextual content that is highly relevant to the target intent. For example, for the target intent of "health consultation," "recent medication history" and "symptom description" are retained, while irrelevant content such as "weather inquiry" is removed.

[0117] Based on the initial weight assignment, the contextual content related to the target intent is filtered through gating operations. Specifically, the gating threshold is dynamically adjusted according to the weight of each historical information to filter out the contextual content that meets the gating conditions:

[0118] Set the initial gating conditions and dynamically adjust the gating threshold based on the weight value of historical information. For example, if the target intent is clear and highly relevant (such as "determine the specific time"), the gating threshold is set higher to retain only the most relevant contextual content; if the target intent is vague (such as "health status consultation"), the gating threshold is set lower to retain more potentially relevant information.

[0119] Apply gating conditions to filter contextual content, retaining only information that meets the gating conditions. For example, when filtering semantic fragments about a user's health status, only contextual content that meets the target intent, such as "blood pressure records" and "medication status," is retained, while other information that does not meet the conditions is eliminated.

[0120] The filtered context content is organized into candidate semantic segments, each of which contains complete turn information, time series features, and semantic features.

[0121] The context content filtered by the gating operation is normalized, including time series sorting and content structuring of the filtered context semantic fragments, and the normalized context content is stored as a reference semantic fragment:

[0122] Sort the filtered semantic fragments in chronological order to ensure the temporal logical coherence of the content.

[0123] Semantic segments are associated with their corresponding timestamps, turn numbers, and contextual semantic features to generate structured data representations. For example, a "symptom description" is stored along with its corresponding conversation time and turn information to form standardized data that is easy to process later.

[0124] The normalized context content is stored as a reference semantic fragment for subsequent intent correction steps. For example, in a health consultation scenario, the reference semantic fragment may include "symptom changes described by the user" and "medication recommendations given by the doctor."

[0125] The multi-task learning structure based on deep neural networks analyzes the reference semantic fragments in parallel, classifies different intents and calculates the score of each intent, including:

[0126] Preprocess the reference semantic fragments according to their contextual relevance and time series features:

[0127] Receive the reference semantic segments filtered and normalized in the previous stage, extract the contextual relevance and time series features of each semantic segment, and ensure that the preprocessed data contains the following:

[0128] Contextual relevance: records the semantic relevance between the reference semantic segment and the target intent. For example, a segment related to health intent contains keywords such as "taking medication" and "symptoms."

[0129] Time series features: Timestamps are marked according to the chronological order of occurrence of semantic segments to preserve temporal continuity in subsequent analysis.

[0130] The extracted contextual relevance information and time series features are bound to the reference semantic fragment to generate structured input data. Each record of the input data should contain the text content of the semantic fragment, semantic relevance information, timestamp, and target intent context label.

[0131] The generated input data is normalized to ensure that the time series features and context relevance of different semantic segments are in a unified numerical range so that they can be received and processed by the deep neural network.

[0132] Construct a multi-task learning structure for a deep neural network, which includes intent classification and intent scoring. The intent classification task is used to identify the intent category corresponding to the reference semantic segment, and the intent scoring task is used to calculate the weight score of each intent:

[0133] Intent classification task module: used to identify the intent category corresponding to the reference semantic segment. For example, health consultation segments are classified as "symptom description" or "treatment advice".

[0134] Intent scoring task module: used to calculate the score of each intent category. The scoring result reflects the matching degree and importance between the semantic segment and the target intent.

[0135] The input layer of the network structure receives preprocessed reference semantic segment data, including semantic segment content, contextual relevance information, and time series features. Each layer of the network processes different features of the input data.

[0136] In the multi-task learning structure, the pre-processed reference semantic segments are processed in parallel, and the intent classification task classifies each semantic segment to determine the intent category to which it belongs:

[0137] In the intent classification task module, a deep neural network classifies the semantic content of a segment, combining contextual relevance and time series features to output the segment's intent category. Classification determines the semantic category based on keywords and contextual information within the segment, for example, classifying "feeling dizzy recently" as "symptom description."

[0138] In the intent scoring task module, the input reference semantic fragments are scored. The scoring criteria include the semantic relevance of the semantic fragments to the target intent, the temporal correlation, and the importance of the fragments in multi-round dialogues.

[0139] To ensure the synchronization of classification results and scoring calculations, network weights are shared between task modules. For example, the same feature extraction layer can be reused for semantic feature extraction, reducing computational overhead and improving processing efficiency.

[0140] In the intent scoring task, each intent category is scored based on the semantic features and contextual relationships of the reference semantic fragment. The scoring is based on the degree of association between the semantic fragment and the target intent and the time series characteristics:

[0141] The intent score is calculated based on multi-dimensional features, including:

[0142] Semantic relevance score: the degree of match between the content of the reference semantic segment and the target intention semantic template;

[0143] Temporal relevance score: the temporal proximity between the time of occurrence of the fragment and the time of execution of the target intention;

[0144] Contextual match score: How closely the fragment content corresponds to the target intent context label.

[0145] In the intent scoring task, each intent category is scored by analyzing the semantic features and contextual relationships of the reference semantic fragment. The scoring process includes the following calculations: First, a correlation score is calculated between the semantic fragment and the target intent based on a semantic matching algorithm. This score measures the degree of match between the fragment's semantic features and the target intent template. Second, a temporal correlation score is calculated based on time series features to measure the proximity between the fragment's occurrence time and the target intent's execution time. Finally, a contextual matching score is calculated based on contextual information to assess the semantic consistency of the fragment with the multi-turn conversation context. Finally, the scores for each dimension are weighted and summed according to predefined weight coefficients to generate a scoring result for each intent category.

[0146] The multi-dimensional scoring results are weighted and integrated according to predefined weight coefficients, and the importance of each dimension can be dynamically adjusted based on the actual needs of the target intent. For example, in a health consultation scenario, time relevance may be more important than context matching.

[0147] Integrate the results of intent classification and intent scoring to generate the classification label and score value for each reference semantic segment:

[0148] Each reference semantic fragment is assigned a classification label, which clearly identifies the intent category to which the fragment belongs; the score corresponding to each fragment reflects its importance and relevance to the target intent.

[0149] The integrated results are stored as structured output data for subsequent intent correction and decision processing steps. For example, in a health consultation application, the integrated results may include snippets and their scores for the "symptom description" category, which can be used to prioritize high-scoring content for recommendation to doctors.

[0150] Based on the results of the classification process and intent scoring, a multi-intent fusion correction operation is performed to output intent information that can distinguish each intent, including:

[0151] Receive the classification labels and scoring values generated by the classification processing and intent scoring results, aggregate the classification labels according to the intent categories, and initially form a classification set of intent categories:

[0152] Obtain the classification processing results and intent scoring results of the reference semantic fragments. The data includes the classification label and scoring value corresponding to each semantic fragment; aggregate the classification labels according to the intent category, and classify the semantic fragments of the same category into a set. For example, the fragments of the "health consultation" category are uniformly stored in the health category set; retain the scoring value of each semantic fragment during the aggregation process to ensure that subsequent sorting can be completed based on the scoring information.

[0153] Based on the intent categories and their ratings in the classification set, sort the intent categories in descending order of ratings and select the intent category with the highest rating as the primary intent:

[0154] The intent categories in the classification set are sorted from high to low by their scores, which indicate the degree of relevance between the semantic fragment and the target intent. The highest-scoring intent category is selected as the primary intent, which reflects the user's core expressive need in the current context. For example, in the "medication consultation" scenario, the highest-scoring intent category might be "medication time." Other intent categories with lower scores but still some relevance are marked as secondary intents for subsequent supplementary processing.

[0155] Perform semantic supplementation on the intent categories with lower scores but certain relevance, and merge them with the main intent to form a comprehensive correction result of multiple intents:

[0156] For the secondary intent category, analyze its semantic relevance to the primary intent. For example, in the health consultation scenario, use "symptom description" as a semantic supplement to the primary intent of "treatment advice"; fuse the supplemented semantic fragments with the primary intent category to form a comprehensive correction result of multiple intentions, ensuring that the corrected content contains both the information of the primary intent and the contextual supplementary information related to it; retain the key information of the secondary intent during the supplementation process to facilitate the fusion and storage of multi-intent content in subsequent steps.

[0157] The semantic information of the main intent and related intents is fused through the multi-intent fusion algorithm to generate intent information that can distinguish each intent:

[0158] Based on the multi-intent fusion algorithm, the semantic information of the primary intent and secondary intent is deeply integrated. The following factors are considered during the fusion process:

[0159] Semantic consistency: Ensure that the primary and secondary intents are consistent in their content. For example, in health consultations, ensure that the information on "medication recommendations" and "medication time" is logically coherent.

[0160] Contextual relevance: Merge contextually relevant content from primary and secondary intents, such as aligning temporal information with semantic content.

[0161] After the fusion is completed, comprehensive semantic information is generated that can distinguish each intent, ensuring that the multi-intent information is independent and complementary to each other.

[0162] The fused and corrected intent information is stored in a structured manner, and a clear classification result that can distinguish between the main intent and related intents is output:

[0163] The results of multi-intent fusion and correction are stored in a structured manner. The stored content includes: classification labels, semantic content and scoring information of each intent category; the output intent information can clearly distinguish between primary and secondary intents. For example, in a user's dialogue request, "medication time" is marked as the primary intent, and "drug name" and "medication frequency" are marked as secondary intents; the structured stored intent information will serve as input data for subsequent decision analysis or interactive response.

[0164] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0165] The above embodiments can be implemented in whole or in part via software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in the embodiments of this application are fully or partially performed. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0166] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0167] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0168] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0169] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0170] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0171] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0172] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

[0173] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for correcting elderly people's conversation intention information based on deep neural network, characterized by: The method includes the following steps: S1: collecting and cleaning multi-round dialogue data containing multiple topics, annotating the multi-round dialogue data and unifying the format; S2: Utilizes the multi-intent recognition sub-model to extract topic features from the annotated multi-round dialogue data and generate multi-intent representation vectors that can represent each intent. S3: Utilizes a hierarchical attention mechanism that dynamically adjusts weights based on user context to perform correlation calculations on the multi-intent representation vectors and filter contextual content related to the target intent. S4: Processes contextual content related to the target intent through gating operations based on multi-round interaction history, retaining reference semantic fragments that are valuable for intent correction. S5: Analyzes the reference semantic fragments in parallel based on a multi-task learning structure based on a deep neural network, classifies different intents, and calculates a score for each intent. S6: Perform multi-intent fusion correction based on the results of the classification process and intent scoring, and output intent information that can distinguish each intent; The multi-intention recognition sub-model is used to extract topic features from the annotated multi-round dialogue data to generate a multi-intention representation vector that can represent each intention. Specifically, the multi-intention recognition sub-model is input into the pre-trained multi-intention recognition sub-model to parse the text content of each round of dialogue and extract semantic information related to the topic features; in the multi-intention recognition sub-model, the semantic analysis module is used to extract features from the text content of each round of dialogue, and the extracted features include semantic features of the topic category and semantic features of the intent category; the embedded vector generation module is used to embed the extracted semantic features to generate an initial vector that can represent the topic category and intent category; the feature fusion module of the multi-intention recognition sub-model is used to fuse the initial vectors generated for each round of dialogue to obtain a multi-intention representation vector that can comprehensively represent multiple intent categories and topic categories; A hierarchical attention mechanism combined with dynamic weight adjustment of user context is used to perform association calculations on multiple intent representation vectors and screen contextual content related to the target intent. Specifically, the multi-intent representation vectors are layered according to predefined attention levels, and each layer assigns initial weights to different dimensional features in the user context information; the weight distribution of each layer of attention is dynamically adjusted according to the historical conversation content and semantic features of the current user context information; in each layer of the attention mechanism, the multi-intent representation vectors are weighted according to the adjusted weight distribution to obtain the weight distribution of different intent categories in each layer; the semantic segments with the strongest correlation with the current target intent are screened out through the weighted results of the multi-layer attention mechanism and the contextual information related to the target intent; the semantic segments with the strongest correlation with the current target intent are uniformly processed to generate a contextual content representation that can represent the target intent; The specific formula for dynamically adjusting the weight distribution of attention in each layer is as follows: ;in, Indicates the adjusted layer attention weights, Indicates the The initial weights of the layers, Indicates user context information and The similarity of the multi-intention representation vector input of the layer, Represents user context information, Indicates the layer multi-intention representation vector input; process the context content related to the target intent through a gating operation based on multi-round interaction history, and retain reference semantic fragments with reference value for intent correction, specifically including: extracting user multi-round interaction history information from the context content related to the target intent, including dialogue rounds, context semantic fragments, and time series features associated with the target intent; perform feature screening on the extracted user multi-round interaction history information, assign an initial weight to each historical information according to its relevance to the target intent, and eliminate historical information whose relevance to the target intent is lower than a preset threshold; based on the assigned initial weight, screen the context content related to the target intent through a gating operation, specifically including dynamically adjusting the gating threshold according to the weight of each historical information to screen out context content that meets the gating conditions; perform normalization on the context content screened out by the gating operation, including time series sorting and content structuring of the screened context semantic fragments, and storing the normalized context content as a reference semantic fragment; A multi-task learning structure based on a deep neural network analyzes the reference semantic segments in parallel, classifies different intents separately and calculates the score of each intent, specifically including: preprocessing the reference semantic segments according to their contextual relevance and time series features; constructing a multi-task learning structure of a deep neural network, which includes an intent classification task and an intent scoring task. The intent classification task is used to identify the intent category corresponding to the reference semantic segment, and the intent scoring task is used to calculate the weight score of each intent; in the multi-task learning structure, the preprocessed reference semantic segments are processed in parallel, and the intent classification task classifies each semantic segment to determine the intent category to which it belongs; in the intent scoring task, each intent category is scored based on the semantic features and contextual relationship of the reference semantic segment, and the scoring is based on the degree of relevance between the semantic segment and the target intent and the time series features; the results of intent classification and intent scoring are integrated to generate a classification label and scoring value for each reference semantic segment; A multi-intent fusion correction operation is performed based on the results of classification processing and intent scoring, and intent information that can distinguish each intent is output, specifically including: receiving classification labels and scoring values generated by the classification processing and intent scoring results, aggregating the classification labels according to the intent categories, and preliminarily forming a classification set of intent categories; based on the intent categories and their scoring values in the classification set, sorting the intent categories in descending order of the scoring values, and screening out the intent category with the highest score as the main intent; performing semantic supplementation processing on the intent categories with lower scores but certain relevance, and fusing them with the main intent to form a comprehensive correction result of multiple intentions; fusing the semantic information of the main intent and related intents through the multi-intention fusion algorithm to generate intent information that can distinguish each intent; storing the fused and corrected intent information in a structured manner, and outputting a clear classification result that can distinguish the main intent from related intents.

2. The method for correcting elderly people's conversation intention information based on deep neural network according to claim 1 is characterized in that: Collect and clean multi-round conversation data containing multiple topics, annotate the multi-round conversation materials and unify the format, specifically including: obtaining multi-round conversation data from multi-channel data sources; cleaning the multi-round conversation data, including removing noise data, incomplete data and meaningless conversation records in the conversation; annotating the cleaned multi-round conversation data using predefined annotation rules, including annotation of topic categories, intent categories and conversation turn information; unifying the format of the annotated multi-round conversation data, including normalization of data fields, unified representation of time series and normalization of conversation turn sorting.

Images