Accompanying robot reliability evaluation method and system based on word vector semantic analysis

Abstract

The invention provides an accompanying robot reliability evaluation method and system based on word vector semantic analysis. The method comprises the steps of obtaining input text data of an accompanying person and response text data of an accompanying robot; encoding the input text data and the response text data by using a pre-trained word vector model to generate an input word vector sequence and a response word vector sequence, and determining semantic difference feature vectors of the input word vector sequence and the response word vector sequence; determining a living habit difference coefficient and a language habit difference coefficient of the accompanying person and the accompanying robot, and determining a reliability correction coefficient of the accompanying robot according to the semantic difference feature vector, the living habit difference coefficient and the language habit difference coefficient; and generating a comprehensive reliability evaluation result of the accompanying robot based on the reliability correction coefficient and a preset reference reliability index. And correction can be performed in combination with living habits and language habits of the accompanying person, so that the real reliability level of the accompanying robot under different habit differences can be reflected.

Description

Technical Field

[0001] The present invention relates to the field of robots, and in particular to a reliability evaluation method and system for a care robot based on word vector semantic analysis. Background Art

[0002] A companion robot is a multifunctional service robot that is mainly used to assist the lives of the elderly, children, the disabled and other groups. It can provide multiple functions such as services, safety monitoring, human-computer interaction and multimedia entertainment.

[0003] In related technologies, due to differences in living habits and language usage, companion robots can easily misinterpret the caregiver's intentions and make incorrect movements, resulting in poor care quality. To improve the quality of companion robots, some robots undergo regular reliability testing to ensure their quality and enhance the user experience. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a companion robot reliability evaluation method and system based on word vector semantic analysis, which solves the problem that the companion robot easily misunderstands the intention of the companion and produces erroneous actions, resulting in low quality of companionship.

[0005] In a first aspect, the present application provides a method for evaluating the reliability of a care robot based on word vector semantic analysis, comprising:

[0006] Obtaining the caregiver's input text data and the care robot's response text data;

[0007] Encode the input text data and the response text data using a pre-trained word vector model to generate an input word vector sequence and a response word vector sequence, and determine a semantic difference feature vector between the input word vector sequence and the response word vector sequence;

[0008] Determining a living habit difference coefficient and a language habit difference coefficient between the accompanying person and the accompanying robot, and determining a reliability correction coefficient of the accompanying robot based on the semantic difference feature vector, the living habit difference coefficient, and the language habit difference coefficient;

[0009] Based on the reliability correction coefficient and the preset benchmark reliability index, a comprehensive reliability evaluation result of the accompanying robot is generated.

[0010] In one embodiment, the word vector model is a BERT model; encoding the input text data and the response text data using the pre-trained word vector model to generate an input word vector sequence and a response word vector sequence specifically includes:

[0011] Merging the input text data and the response text data through the classification identifier and the separation identifier to generate a joint input sequence;

[0012] Processing the combined input sequence through the embedding layer of the BERT model to generate a word vector matrix; each row in the word vector matrix corresponds to a context-dependent vector of a subword;

[0013] The word vector matrix boundary is positioned based on the positions of the classification identifier and the separation identifier, thereby separating the input word vector sequence and the response word vector sequence.

[0014] In one embodiment, the merging and processing the input text data and the response text data to generate a combined input sequence specifically includes:

[0015] Adding a classification identifier to the headers of the input text data and the response text data;

[0016] Adding a first separator between the input text data and the response text data;

[0017] Adding a second separator mark at the end of the input text data and the response text data;

[0018] Based on the classification identifier, the first separation identifier, and the second separation identifier, the input text data and the response text data are merged to form a joint input sequence.

[0019] In one embodiment, locating the boundaries of the word vector matrix based on the positions of the classification identifier and the separation identifier, thereby separating the input word vector sequence and the response word vector sequence, specifically includes:

[0020] Separate the input word vector sequence from the index position after the classification identifier to the index position before the first separation identifier in the index position of the joint input sequence;

[0021] In the index positions of the joint input sequence, the response word vector sequence is separated from the index position after the first separation identifier to the index position before the second separation identifier.

[0022] In one embodiment, determining the semantic difference feature vectors of the input word vector sequence and the response word vector sequence specifically includes:

[0023] The input word vector sequence and the response word vector sequence are used as input to generate a bidirectional context feature vector through a Transformer model; the Transformer model includes a query space, a key space, and a value space, and the query space, the key space, and the value space correspond to a query weight matrix, a key weight matrix, and a value weight matrix, respectively;

[0024] Taking the global semantic vector corresponding to the response word vector sequence in the bidirectional context feature vector to construct a query vector, taking the local feature vector corresponding to the input word vector sequence in the bidirectional context feature vector to construct a key vector, and generating a difference weight distribution by element-by-element subtraction of the query vector and the key vector; the query vector is obtained by multiplying the global semantic vector and the query weight matrix; the key vector is obtained by multiplying the local feature vector and the key weight matrix;

[0025] The semantic difference feature vector is generated based on the difference weight distribution and weighted summation of the value vector; the value vector is obtained by multiplying the input word vector sequence by the value weight matrix.

[0026] In one embodiment, the semantic difference feature vector is calculated and determined by the following formula:

[0027]

[0028] in, is the semantic difference feature vector; Indicates the The query vector of the sequence; Indicates the The key vector of the sequence; is the sequence length of the bidirectional context feature vector; Indicates the The dimension of the key vector of the sequence; is a value vector.

[0029] In one embodiment, determining the difference coefficient of living habits and language habits between the accompanying person and the accompanying robot specifically includes:

[0030] Obtaining the living habit difference coefficient based on historical interaction data between the caregiver and the care robot through a hierarchical analysis method;

[0031] The similarity between the user's pronunciation data and the preset dialect database is calculated to obtain the language habit difference coefficient.

[0032] In one embodiment, the method of obtaining the living habit difference coefficient based on the historical interaction data between the caregiver and the care robot by using the hierarchical analysis method specifically includes:

[0033] Collecting historical interaction data between the caregiver and the care robot; the historical interaction data includes the caregiver's living habit label and the robot's response accuracy;

[0034] Constructing a life habit evaluation index system based on the historical interaction data; the life habit evaluation index system includes multiple first-level indicators, each of which has corresponding second-level sub-indicators;

[0035] The weight vector of the first-level indicator is calculated by the hierarchical analysis method, and the entropy weight method weight of the second-level sub-indicator is combined to generate the living habit difference coefficient.

[0036] In one embodiment, after generating a comprehensive reliability evaluation result of the accompanying robot based on the reliability correction coefficient and the preset benchmark reliability index, the method further includes:

[0037] Determining a confidence score for the comprehensive reliability evaluation result;

[0038] If the confidence score is less than a preset confidence threshold, the accompanying robot is controlled to enter an adaptive learning mode; in the adaptive learning mode, the accompanying robot can optimize the word vector model according to the input text data obtained in real time.

[0039] In a second aspect, the present application provides a companion robot reliability evaluation system based on word vector semantic analysis, comprising a processor and a memory; wherein the memory stores a computer program, and the computer program is used to be loaded by the processor and execute the companion robot reliability evaluation method based on word vector semantic analysis as described in any one of the first aspects.

[0040] In the accompanying robot reliability evaluation method and system based on word vector semantic analysis of this embodiment, the word vector model can capture the semantic information of words, so that the input text data and response text data are converted into quantifiable vector form, so as to generate semantic difference feature vectors to focus on the semantic deviation between input and response, and make corrections based on the living habits and language habits of the accompanying person, so that the generated reliability evaluation results can reflect the true reliability level of the accompanying robot under different habit differences, avoiding evaluation distortion caused by ignoring habit differences. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 A flowchart of a method for evaluating the reliability of a companion robot based on word vector semantic analysis is provided for one embodiment of the present application.

[0043] Figure 2 A schematic structural diagram of an electronic device provided in accordance with an embodiment of the present application. DETAILED DESCRIPTION

[0044] Specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the described embodiments are merely some, and not all, of the embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the description of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0045] In the description of the present invention, unless otherwise specified or limited, the terms "disposed," "installed," and "connected" should be understood broadly. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; and direct or indirect connections through an intermediary. Those skilled in the art will understand the specific meanings of these terms based on the specific circumstances.

[0046] The directions or positional relationships indicated by terms such as "upper", "lower", "left", "right", "front", "back", "top", "bottom", "inside" and "outside" are based on the directions or positional relationships shown in the accompanying drawings, or are the directions or positional relationships in which the inventive product is usually placed when in use. They are only for the convenience and simplification of description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operate in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0047] The terms "first," "second," "third," etc. are merely used to distinguish elements of similar nature and do not indicate or imply relative importance or a particular order.

[0048] The terms "comprises," "comprising," or any other variations thereof, are intended to cover a non-exclusive inclusion of elements other than the listed elements and may also include additional elements not specifically listed.

[0049] like Figure 1 As shown, this embodiment provides a method for evaluating the reliability of a care robot based on word vector semantic analysis, including:

[0050] Step S100: Acquire the input text data of the accompanying person and the response text data of the accompanying robot;

[0051] Step S200: Encode the input text data and the response text data using a pre-trained word vector model to generate an input word vector sequence and a response word vector sequence, and determine a semantic difference feature vector between the input word vector sequence and the response word vector sequence;

[0052] Step S300: determining a living habit difference coefficient and a language habit difference coefficient between the accompanying person and the accompanying robot, and determining a reliability correction coefficient of the accompanying robot based on the semantic difference feature vector, the living habit difference coefficient, and the language habit difference coefficient;

[0053] Step S400: generating a comprehensive reliability evaluation result of the accompanying robot based on the reliability correction coefficient and the preset benchmark reliability index.

[0054] In the companion robot reliability evaluation method based on word vector semantic analysis of this embodiment, the word vector model can capture the semantic information of words, so that the input text data and response text data are converted into quantifiable vector form, so as to generate semantic difference feature vectors to focus on the semantic deviation between input and response, and make corrections in combination with the living habits and language habits of the companion, so that the generated reliability evaluation results can reflect the true reliability level of the companion robot under different habit differences, avoiding the evaluation distortion caused by ignoring habit differences.

[0055] Step S100: Acquire the input text data of the accompanying person and the response text data of the accompanying robot.

[0056] A caregiver is an individual who requires care services, including the elderly, children, and people with disabilities. Input text data refers to commands or questions sent by a caregiver to the care robot via voice or text, reflecting the user's immediate needs, such as "Please open the curtains" or "I need a reminder to take my medicine."

[0057] Response text data refers to the response content generated by the accompanying robot based on the input data. It is used to reflect the robot's understanding and execution capabilities, such as "Opening the curtains for you" or "A medication reminder has been set."

[0058] The voice recognition module or input device can capture the caregiver's instructions in real time and convert them into input text data. The robot can also generate response text data by recording the actions or language it responds to these instructions. Furthermore, the input and response text data cover a variety of life scenarios, such as daily care, emergency assistance, and entertainment interactions. The system also records the timestamps of the relevant terms in the input and response text data for each scenario, ensuring that the temporal relationship between the input and response text data corresponds.

[0059] Step S200: Use a pre-trained word vector model to encode the input text data and the response text data, generate an input word vector sequence and a response word vector sequence, and determine the semantic difference feature vectors of the input word vector sequence and the response word vector sequence.

[0060] The word vector model is a model that has been trained on a large-scale corpus and can map words into numerical vectors, such as models such as Word2Vec, GloVe or BERT. The BERT model is preferred in this embodiment. The input word vector sequence is a sequence formed by converting each word in the input text into a fixed-dimensional vector. For example, "open the curtains" may be encoded as [[0.1, 0.2, ...], [0.3, 0.4, ...]]. The response word vector sequence is a vector sequence generated after the response text is processed by the same model. The semantic difference feature vector is a vector that quantifies the degree of semantic deviation between the input and response texts.

[0061] In one embodiment, encoding the input text data and the response text data using a pre-trained word vector model to generate an input word vector sequence and a response word vector sequence specifically includes:

[0062] Step S201: The input text data and the response text data are combined and processed by using a classification identifier and a separation identifier to generate a joint input sequence.

[0063] The classification identifier is a specific marking symbol used to identify the starting position of a text sequence. It can be obtained by adding it to the starting end of the input text data during the merging process. The classification identifier of this embodiment includes one or more of the starting markers predefined by the BERT model.

[0064] The separator is a specific marking symbol used to distinguish text paragraphs from different sources. It can be obtained by inserting it at the junction of the input text data and the response text data during the merging process. The separator in this embodiment includes one or more paragraph separators predefined by the BERT model.

[0065] The combined input sequence is a composite sequence formed by concatenating input text data and response text data through a classification identifier and a separation identifier.

[0066] In one embodiment, the merging and processing of the input text data and the response text data to generate a joint input sequence specifically includes: adding a classification identifier to the header of the input text data and the response text data; adding a first separator identifier between the input text data and the response text data; adding a second separator identifier to the tail of the input text data and the response text data; and merging the input text data and the response text data with the classification identifier, the first separator identifier, and the second separator identifier to form a joint input sequence.

[0067] Among them, the classification identifier is specifically the [CLS] tag, which is used to identify the starting position of the input text data and the response text data, so as to facilitate the extraction of global semantic features in subsequent model processing.

[0068] The first delimiter identifier is a predefined symbol used to distinguish the boundary between the input text and the response text content, which can be implemented by a symbol different from the classification identifier. Exemplarily, the first delimiter identifier may include the [SEP] token.

[0069] The second delimiter identifier can be a predefined symbol used to mark the end position of the entire combined input sequence, which can be implemented by the same or different symbols as the first delimiter identifier. Exemplarily, the second delimiter identifier may include the [SEP] token.

[0070] The text data with added identifiers is combined to form a combined input sequence. Exemplarily, it can be implemented by concatenating the classification identifier, the input text data, the first delimiter identifier, the response text data, and the second delimiter identifier in sequence, for example, forming a structure similar to "[CLS] input content [SEP] response content [SEP]".

[0071] By adding a classification identifier to the head of the input text and the response text to extract global features, inserting the first delimiter identifier between them to distinguish the content boundary, adding the second delimiter identifier at the end to clarify the sequence termination position, and combining the above identifiers with the text data to form a structured combined input sequence, the technical effects of strengthening text boundary recognition, improving the model adaptability and fault tolerance ability through double delimiter identifiers can be achieved. In this embodiment, the cooperation of the double delimiter identifiers avoids the content confusion problem in multi-round interaction or complex instruction scenarios, ensures that the BERT model can generate a stable word vector matrix according to the pre-training format, and reduces the risk of vector drift or truncation caused by abnormal sequence formats, reducing the calculation error of the semantic difference feature vectors, thereby improving the calculation accuracy of the reliability correction coefficient.

[0072] Step S202: Process the combined input sequence through the embedding layer of the BERT model to generate a word vector matrix; each row in the word vector matrix corresponds to a context-related vector of a sub-word.

[0073] The embedding layer of the BERT model is a superimposed module that includes word embedding, position embedding, and segment embedding, and is used to convert the text sequence into a numerical vector representation. Word embedding can be achieved by mapping sub-words to vectors of a fixed dimension, position embedding can be achieved by encoding the order information of sub-words in the sequence, and segment embedding can be achieved by marking the attribution of different text paragraphs. The word vector matrix is a two-dimensional matrix generated after being processed by the embedding layer, the number of rows of which corresponds to the number of sub-word units, and the number of columns corresponds to the preset vector dimension. Exemplarily, each element value of the word vector matrix can reflect the comprehensive semantic and position features of sub-words in a specific context. For example, the vector of "take" in "Please take medicine" is different from the vector in "Pick up the book".

[0074] In one embodiment, locating the boundaries of the word vector matrix based on the positions of the classification identifier and the separation identifier, thereby separating the input word vector sequence and the response word vector sequence, specifically includes:

[0075] Step S2021: Separate the input word vector sequence from the index position after the classification identifier to the index position before the first separation identifier in the index position of the combined input sequence;

[0076] Step S2022: Separate the response word vector sequence from the index position after the first separation marker to the index position before the second separation marker in the index position of the joint input sequence.

[0077] In step S2021 and step S2022, by determining the absolute positions of the classification identifier and the separation identifier, the vector range of the input and response texts can be delineated, and the risk of fuzzy matching based on text content is avoided by the absolute index rule, and the accuracy of boundary positioning is maintained. Exemplarily, the input word vector sequence and the response word vector sequence can be separated by the following steps: first, the index value of the classification identifier in the joint input sequence is located (for example, the index of [CLS] is 0); secondly, the index value of the first separation identifier in the joint input sequence is determined (for example, the index of the first [SEP] is 4); then the starting index of the input word vector sequence is set to the position after the classification identifier (index 1), and the ending index is set to the position before the first separation identifier (index 3), thereby extracting the vector set within the interval. For the response word vector sequence, by locating the index value of the second separation identifier (such as index 9), the starting index is set to the position after the first separation identifier (index 5), and the ending index is set to the position before the second separation identifier (index 8), thereby completing the vector extraction of the response text.

[0078] By using the absolute index positions of classification identifiers and separation identifiers to define the vector boundaries of the input and response texts, and combining them with strict start and end rules, the extraction of complete semantic units can be achieved. At the same time, the second separation identifier is used as the termination anchor to prevent the response sequence from over-expansion. The mathematical certainty of the index rule significantly reduces the vector truncation deviation caused by identifier recognition errors, and can still be fault-tolerant through the relative positions of other identifiers when the classification identifier is lost; and ensure that the complete semantic units of the input and response texts are retained. For example, all relevant vectors can still be fully captured in response texts containing multiple clauses or corrective supplements; and experimental verification reduces the correspondence error between the word vector sequence and the original text, thereby improving the accuracy of subsequent semantic difference analysis and the credibility of the reliability correction coefficient.

[0079] Step S203: Locate the boundaries of the word vector matrix based on the positions of the classification identifier and the separation identifier, thereby separating the input word vector sequence and the response word vector sequence.

[0080] Boundary localization refers to determining the vector subsets corresponding to the input text and the response text by parsing the position indices of the classification identifier and the delimiter identifier in the word vector matrix. Exemplarily, if the delimiter identifier is located in the 5th row, the first 5 rows correspond to the input word vector sequence, and the subsequent rows correspond to the response word vector sequence. The index positions of the classification identifier and the delimiter identifier can be identified by traversing the row identifiers of the word vector matrix, so as to ensure the correspondence between the separated word vector sequences and the original text.

[0081] In a specific embodiment, the input text and the response text can be merged into a unified sequence through the classification identifier and the delimiter identifier. For example, "I'm hungry" and "Lunch has been prepared for you" are merged into "[CLS]I'm hungry[SEP]Lunch has been prepared for you". Then the merged sequence is input into the embedding layer of BERT, and word embeddings, position embeddings, and segment embeddings are stacked to generate a word vector matrix. For example, the vector of the word "hungry" is adjusted in value due to the position relationship of the previous word "I" and the subsequent word "le". Then, the vector subsets of the input and the response are divided according to the index value of the delimiter identifier. For example, when [SEP] is located in the 3rd row, the first 3 rows form the input word vector sequence. Finally, the separated sequence is output for subsequent analysis. For example, the input sequence contains the context vector of "I'm hungry", and the response sequence contains the vector of "Lunch has been prepared for you".

[0082] In steps S201 - S203, by using the embedding layer of the BERT model to generate a word vector matrix containing context information, using the classification identifier and the delimiter identifier to clearly distinguish the boundaries between the input and the response texts, and combining the boundary localization technology to accurately segment the vector subsets, it is possible to improve the context semantic capture ability of word vectors for fuzzy instructions or dialect expressions, avoid vector contamination caused by sequence mixing, and at the same time provide a reliable basis for subsequent semantic difference calculation. It solves the problem of fuzzy semantic representation of traditional word vector models in scenarios with short texts or strong context dependence. For example, when processing dialect expressions such as "Open the window a bit", the BERT model can map "Open a bit" and "Open" to similar vectors based on the context, while traditional models may make misjudgments due to not having seen the vocabulary usage in a specific context. Ultimately, these improvements enable the semantic difference feature vectors to more accurately reflect the true degree of the robot's understanding deviation, thereby improving the calculation accuracy of the reliability correction coefficient.

[0083] In one of the embodiments, determining the semantic difference feature vectors of the input word vector sequence and the response word vector sequence specifically includes:

[0084] Step S204: Using the input word vector sequence and the response word vector sequence as inputs, generate bidirectional context feature vectors through a Transformer model; the Transformer model includes a query space, a key space, and a value space, and the query space, the key space, and the value space respectively correspond to a query weight matrix, a key weight matrix, and a value weight matrix;

[0085] The Transformer model is a deep neural network architecture based on the self-attention mechanism, which generates feature vectors by parallel computing the context dependencies at different positions. Its core modules include a multi-head attention layer and a feed-forward network. Exemplarily, this model can simultaneously capture the relevance between any two words in the input sequence. For example, when analyzing the response of "Please get medicine" and "Getting medicine now", it can identify the semantic association between "get" and "fetch" in different contexts.

[0086] The query space, the key space, and the value space are linear transformation spaces defined by the query weight matrix, the key weight matrix, and the value weight matrix respectively, which are used to map input vectors to achieve feature comparison in different dimensions. The query weight matrix is the linear transformation parameter that converts the global semantic vector of the response into a query vector. For example, the global vector of the response forms a representation in the query space after matrix multiplication. The key weight matrix is the linear transformation parameter that converts the local feature vector of the input into a key vector. For example, "get medicine" in the input text forms a feature representation in the key space. The value weight matrix is the linear transformation parameter that converts the original input word vector into a value vector. For example, "medicine" retains specific semantic features in the value space.

[0087] The bidirectional context feature vectors are a set of feature vectors containing global and local semantic information output by the Transformer model, which can be generated by calculating the association weights between each word and other words in the sequence through the self-attention mechanism.

[0088] Step S205: Take the global semantic vector corresponding to the response word vector sequence in the bidirectional context feature vectors to construct a query vector, take the local feature vector corresponding to the input word vector sequence in the bidirectional context feature vectors to construct a key vector, and generate a difference weight distribution through the element-wise subtraction of the query vector and the key vector; the query vector is obtained by multiplying the global semantic vector by the query weight matrix; the key vector is obtained by multiplying the local feature vector by the key weight matrix.

[0089] The global semantic vector is a comprehensive representation obtained by aggregating the overall information of the response word vector sequence; Exemplarily, the global vector of the response "Lunch has been prepared for you" can reflect the core intention of "providing catering services".

[0090] Local feature vectors are fine-grained feature representations of each word or each segment in the input word vector sequence. For example, the “hungry” in “I am hungry” retains the direct expression of the hunger state.

[0091] The query vector can be constructed by multiplying the global semantic vector with the query weight matrix. For example, the global vector responding to "closed window" is transformed into a semantic representation in the query space after matrix transformation.

[0092] The key vector construction can be achieved by multiplying the local feature vector with the key weight matrix, for example, the word-by-word vector of the input "window" is transformed to form the features in the key space.

[0093] Element-by-element subtraction calculates the difference between the query vector and the key vector along the corresponding dimension, generating a weighted distribution that quantifies the degree of semantic deviation. For example, if the input word "hungry" and the response word "lunch" differ significantly along a certain dimension, the corresponding position will have a higher weight.

[0094] Step S206: Generate the semantic difference feature vector based on the difference weight distribution and weighted sum of the value vector; the value vector is obtained by multiplying the input word vector sequence by the value weight matrix.

[0095] The value vector is a vector carrying the original semantic information generated after the input word vector sequence is transformed through the value weight matrix. For example, the value vector corresponding to "open window" retains the specific characteristics of the "open action".

[0096] Weighted summation combines value vectors based on their difference weight distribution. For example, the value vectors corresponding to "medicine" and "medicine extraction," which have higher difference weights, contribute more to the final difference feature vector. Dynamically adjusting the contribution of features at each position through the difference weight distribution avoids the averaging flaw of traditional mean methods.

[0097] In one embodiment, the semantic difference feature vector is calculated and determined by the following formula:

[0098]

[0099] in, is the semantic difference feature vector; Indicates the The query vector of the sequence; Indicates the The key vector of the sequence; is the sequence length of the bidirectional context feature vector; Indicates the The dimension of the key vector of the sequence; is a value vector.

[0100] The semantic difference feature vector is a final vector that comprehensively reflects the degree of difference between the input and response text in terms of semantics, context, and local features. Its dimensionality is consistent with the vector space of the value vector, and each element represents the intensity of the deviation in a specific semantic dimension. For example, when there is a significant difference in the "open state" dimension between the input "window open" and the response "window closed" in the "open state" dimension, the corresponding element value of the semantic difference feature vector will be higher.

[0101] It can be generated by inputting the response global semantic vector into the query weight matrix for matrix multiplication , and input the local feature vector into the key weight matrix to generate , and then the input word vector sequence is input into the value weight matrix to generate the value vector V. For example, after the global vector 768 dimensions is transformed by the 768×64 matrix The key vector is also 64-dimensional, and the value vector dimension can be 512. The difference weight distribution can be calculated by calculating for each position i (1≤i≤L), and The element-by-element difference of The difference vector of all position difference vectors is then concatenated to form Matrix implementation, for example and The difference is [0.1,0.1].

[0102] In steps S204-S206, the Transformer's bidirectional attention mechanism ensures that the difference analysis takes into account the complete contextual information, such as identifying key related words in multi-step instructions; secondly, the separation of global and local features enables the difference analysis to take into account both overall intentions and detailed deviations, such as distinguishing intent compliance from specific preference deviations in the responses of "play music" and "classical music has been played"; thirdly, the dynamic weight allocation mechanism accurately locates the difference position through element-by-element subtraction, such as distinguishing the reference difference between "open the window" and "opened the window" in dialect scenarios; finally, the value vector retains the original semantic features to ensure the fidelity of the difference feature vector, such as strengthening the semantic distinction between "medicine" and "book" in the reference error scenario.

[0103] By inputting the input and response sequences into the Transformer model to generate a bidirectional context feature vector, the query weight matrix and the key weight matrix are used to construct the comparison vectors of global and local features respectively, and the degree of semantic deviation is quantified through element-by-element subtraction. Finally, the difference weight distribution and the value vector are combined to generate a difference feature vector. This can reduce the representation error of the semantic difference feature vector, improve the calculation accuracy of the reliability correction coefficient, and control the reliability evaluation error of the accompanying robot in complex instruction scenarios within 5%, effectively reducing the risk of misjudgment due to semantic misunderstanding.

[0104] Step S300: Determine the living habit difference coefficient and language habit difference coefficient between the companion and the companion robot, and determine the reliability correction coefficient of the companion robot based on the semantic difference feature vector, the living habit difference coefficient and the language habit difference coefficient.

[0105] The lifestyle difference coefficient is an indicator used to quantify the differences in daily behavior patterns between users and robots, such as sleep and rest time, dietary preferences, or operating habits. It can be calculated through questionnaires, historical behavior data statistics, or sensor records (such as activity frequency).

[0106] The language habit difference coefficient is an indicator used to reflect the differences in expression between users and robots, such as dialect usage, terminology preference, or sentence structure. For example, the elderly may prefer to use colloquial expressions while the robot defaults to standard Mandarin.

[0107] The reliability correction coefficient can be an adjustment factor generated by weighted fusion of the above-mentioned living habit difference coefficient, language habit difference coefficient and semantic difference feature vector. For example, if the user is accustomed to using dialects, the correction coefficient may amplify the impact of language differences on the evaluation.

[0108] In one embodiment, determining the difference coefficient of living habits and language habits between the accompanying person and the accompanying robot specifically includes:

[0109] Step S301: Obtain the living habit difference coefficient based on the historical interaction data between the caregiver and the care robot through the hierarchical analysis method.

[0110] The analytic hierarchy process (AHP) converts subjective judgments into quantifiable weight values ​​by constructing a judgment matrix and a hierarchical structure. Its core includes establishing a target layer, a criterion layer, and a solution layer, calculating eigenvectors, and performing consistency checks. For example, the AHP can include allocating weights to sub-dimensions such as work and rest patterns, operating preferences, and eating habits.

[0111] In one embodiment, the lifestyle difference coefficient is obtained based on the historical interaction data between the caregiver and the care robot through the hierarchical analysis method, which specifically includes: collecting the historical interaction data between the caregiver and the care robot; the historical interaction data includes the caregiver's lifestyle label and the robot's response accuracy; constructing a lifestyle habit evaluation index system based on the historical interaction data; the lifestyle habit evaluation index system includes multiple first-level indicators, and each first-level indicator has a corresponding second-level sub-indicator; calculating the weight vector of the first-level indicator through the hierarchical analysis method, and combining the entropy weight method weights of the second-level sub-indicators to generate the lifestyle difference coefficient.

[0112] Among them, historical interaction data is a structured data set that records the interaction behavior between users and robots. It is obtained by parsing and labeling historical conversation records, operation logs and environmental sensor data. For example, the lifestyle labels include quantitative indicators that describe user behavior patterns, such as regularity of work and rest, command time distribution, and voice misrecognition rate.

[0113] The response accuracy rate is the probability of a robot successfully completing a user command in a specific interaction scenario. It is calculated by counting the ratio of the number of successful responses to the total number of interactions, for example, the proportion of correct responses given by the robot to commands issued by the user at night.

[0114] The lifestyle evaluation index system is a multi-level evaluation framework for quantitatively assessing the differences between user and robot behaviors. It achieves systematic analysis by decomposing user behavior characteristics into measurable indicator levels. The first-level indicators are abstract dimensions that reflect the core lifestyle characteristics of users, such as regularity of work and rest, and matching of operating habits. The second-level sub-indicators are detailed parameters that specifically describe the first-level indicators. For example, under regularity of work and rest, they include frequency of nighttime interactions and fluctuation rate of instruction time. In a specific embodiment, the indicator system can be constructed by combining interviews with domain experts with data cluster analysis to ensure that the main dimensions of user behavior are covered.

[0115] The first-level indicator weight vectors calculated using the Analytic Hierarchy Process (AHP) are subjective weight distribution results determined through expert ratings or user surveys. The generation process involves constructing a matrix for judging the relative importance of indicators, calculating eigenvectors, and verifying the rationality of the weights through consistency tests. The entropy weight method uses objective weights calculated based on the degree of variability in the data for the second-level sub-indicators. The calculation process involves calculating the information entropy for each sub-indicator dataset and then assigning weights based on the entropy value. For example, greater volatility in a sub-indicator's data indicates lower information entropy, and a higher weight distribution ratio is assigned to it.

[0116] In a specific embodiment, when determining the coefficient of difference in living habits, the historical interaction data is first labeled to extract the characteristics of user living habits and the robot response performance parameters; secondly, an evaluation system containing multi-level indicators is constructed according to business needs, for example, "operation habit matching" is set as a first-level indicator, which includes second-level indicators such as gesture recognition accuracy and voice command false touch rate; again, the subjective weight of each first-level indicator is determined through the hierarchical analysis method, for example, the weight ratio of "regularity of work and rest" and "consistency of environmental preferences" is determined through pairwise comparison by experts; then, the entropy weight of the second-level sub-indicator data under each first-level indicator is calculated, for example, the information entropy of the "night interaction frequency" data is calculated and its weight coefficient is determined; finally, the hierarchical analysis method weight of the first-level indicator is multiplied by the entropy weight method weight of the second-level sub-indicator to obtain the composite weight coefficient of each sub-indicator, and then the living habit difference coefficient is generated through weighted calculation.

[0117] Furthermore, the collaborative optimization of subjective and objective weights improves the adaptability of evaluation results to dynamic changes in user behavior. By collecting historical interaction data containing lifestyle labels and response accuracy, an evaluation system consisting of multiple indicators is constructed. The analytic hierarchy process (AHP) is used to determine the subjective weights of the primary indicators. The entropy weight method is then used to calculate the objective weights of the secondary sub-indicators based on the degree of data variation. Ultimately, a composite weight coefficient is generated to quantify differences. Subjective weights ensure the dominant role of expert experience in key indicators, while a data-driven entropy weight method is used to dynamically optimize the weight distribution of sub-indicators, thereby enhancing the comprehensiveness and accuracy of the evaluation results. Through the composite calculation mechanism of weight coefficients, the quantification of lifestyle differences accurately reflects the actual deviation between user behavior and robot responses. When user interaction patterns change significantly, the system automatically increases the weights of relevant indicators using the entropy weight method, avoiding the evaluation distortion caused by the rigid weighting of traditional single methods. This effectively identifies the impact of implicit differences, such as the "nighttime interaction failure rate," on reliability evaluation, thereby supporting the robot's continued provision of precise service in scenarios with dynamically changing user behavior.

[0118] Step S302: Calculate the similarity between the user's pronunciation data and the preset dialect database, thereby obtaining the language habit difference coefficient.

[0119] The pronunciation data is the acoustic feature parameters in the user's voice interaction, including fundamental frequency, speech rate, phoneme duration, etc. For example, the pronunciation data may include the acoustic feature record of "Air Conditioning" in Cantonese.

[0120] A dialect database is a collection of speech or text samples in different dialects. For example, the dialect database may include an acoustic template for "windowing" in Wu or Minnan dialects.

[0121] When determining the language habit difference coefficient by calculating the similarity between pronunciation data and a dialect database, the system first collects the user's voice signal and extracts acoustic features. The feature vector is then dynamically time-warped or cosine-matched with the dialect database template. The difference is then graded based on the similarity score and a correction coefficient is mapped. Finally, the database and correction coefficient are continuously updated to adapt to changes in user habits. This allows for precise identification of dialect differences. For example, if a user pronounces the word "medicine" in Wu dialect, acoustic feature matching can be used to avoid misidentification. A dynamic update mechanism also allows the coefficient to adjust as the user's language habits change. For example, as the user's use of Mandarin increases, the correction coefficient approaches the baseline value.

[0122] In steps S301 and S302, the differences in living habits are decomposed into quantifiable sub-dimensions by using the hierarchical analysis method and combined with weight calculation. At the same time, accurate identification of language habit differences is achieved based on the similarity matching of acoustic features and the dialect database, and the adaptability of coefficient calculation is improved through a dynamic update mechanism. This can improve the accuracy of difference coefficient calculation, reduce the risk of reliability misjudgment due to differences in dialects or behavioral patterns, and enhance the long-term adaptability of the system.

[0123] Step S400: generating a comprehensive reliability evaluation result of the accompanying robot based on the reliability correction coefficient and the preset benchmark reliability index.

[0124] Baseline reliability metrics are general evaluation criteria that don't account for user differences. Examples include response accuracy, task completion time, or failure rate. These metrics are typically determined through laboratory testing or historical data. Comprehensive reliability evaluation results are the final score obtained by weighting the baseline metrics using correction factors. For example, a baseline accuracy of 90% might be adjusted to 85% to account for differences in user habits.

[0125] The correction factor can be combined with the baseline metric through mathematical operations. For example, if the baseline reliability metric is "response accuracy," the overall evaluation result = baseline accuracy x reliability correction factor. A negative correction factor indicates a decrease in reliability due to variance, resulting in a lower-than-baseline value; a negative correction factor may indicate an improvement. By providing dynamic, personalized reliability assessments, manufacturers can be guided to optimize their products, even when the scores of the same robot vary significantly across user groups.

[0126] In one embodiment, after generating the comprehensive reliability evaluation result of the companion robot based on the reliability correction coefficient and the preset benchmark reliability index, it also includes: determining the confidence score of the comprehensive reliability evaluation result; if the confidence score is less than the preset confidence threshold, controlling the companion robot to enter the adaptive learning mode; in the adaptive learning mode, the companion robot can optimize the word vector model according to the input text data obtained in real time.

[0127] The confidence score is a numerical value that quantifies the trustworthiness of the comprehensive reliability evaluation results. It generates an output probability or model uncertainty estimate based on the statistical model. This score can be obtained by analyzing uncertainty indicators such as the variance of the semantic difference feature vector and the dispersion of the attention weight distribution of the Transformer model. For example, when the user instruction "Help me get that" is semantically ambiguous due to lack of context, the corresponding confidence score may be lower than the confidence score of the user's explicit instruction "Open window".

[0128] The confidence threshold is a preset critical value used to determine the reliability boundary of the reliability evaluation result. Its setting methods include but are not limited to fixed values or dynamic adjustment mechanisms. Exemplarily, if the threshold is set to 0.7, when the confidence score is lower than this value, the system will trigger the companion robot to enter the adaptive learning mode.

[0129] The adaptive learning mode is a special working mode that the robot enters, allowing the model parameters to be dynamically adjusted through real-time interaction data rather than relying entirely on pre-trained parameters. Exemplarily, in this mode, the robot can collect user dialect pronunciation data and update the embedding layer parameters of the word vector model.

[0130] The input text data obtained in real time refers to the instructions or conversation content instantaneously issued by the user during the current interaction process. Its acquisition channels include the speech recognition module or the text input interface. Exemplarily, the sentence "Turn off the light" spoken by the user in dialect belongs to such data.

[0131] The update of the model parameters to optimize the word vector model can be achieved through online learning or fine-tuning techniques. Its specific forms include adjusting the embedding layer weights of the BERT model or the attention mechanism parameters of the Transformer. Exemplarily, the optimized model can improve the semantic vector similarity between "Turn off" and "Switch off".

[0132] The system can conduct quantitative evaluation by extracting indicators such as the variance of the semantic difference feature vectors and the historical interaction data coverage rate, and combining a weighted formula (such as confidence = 0.7×variance + 0.3×coverage rate). Through the synergistic effect of confidence calculation and threshold comparison, the system can judge the credibility of the reliability evaluation result. When the calculation result is lower than the threshold, the system will trigger the mode switching mechanism, causing the robot to pause the normal response and start the adaptive learning mode. In this mode, the robot fuses the real-time input text data with the historical data through incremental training, updates the model parameters using algorithms such as mini-batch gradient descent, and can also actively request user feedback to optimize semantic understanding. For example, when the user issues a vague instruction "Open the window", the system not only performs the window-opening action but also records the pronunciation features and asks the user if they are satisfied with the current operation, thereby adjusting the model parameters. Finally, when the confidence returns above the threshold or after completing the predetermined learning cycle, the system will exit the adaptive mode and solidify the optimized model parameters, recalculate the reliability evaluation result.

[0133] By dynamically evaluating confidence scores and triggering adaptive learning modes, combined with real-time data-driven model optimization strategies, the system can dynamically adjust its operating mode based on the uncertainty of interaction scenarios. By quantifying the credibility boundaries of reliability evaluation results, resource-intensive learning processes are initiated only when necessary, thereby improving the model's adaptability to individual differences while ensuring response efficiency. The system continuously collects data on user-specific dialect expressions, semantic deviation cases, and other data to achieve progressive optimization of model parameters, effectively reducing the accumulation of semantic misunderstandings in long-term service. At the same time, active learning strategies enhance the interactive feedback loop between users and the system, making the model evolution process transparent and thereby enhancing user trust.

[0134] To summarize, in the companion robot reliability evaluation method based on word vector semantic analysis provided in this embodiment, the word vector model is used to convert unstructured text into a computable vector form to accurately capture semantic deviations; the benchmark indicators are corrected in combination with the user individual difference coefficient to solve the distortion problem caused by ignoring user characteristics in traditional evaluation methods; the evaluation results are dynamically adjusted to reflect the actual performance in different scenarios, reducing the frequency of return to factory inspection and maintenance costs, while improving users' trust and satisfaction with robot services.

[0135] Based on the same inventive concept as the above embodiment, this embodiment also provides a companion robot reliability evaluation system based on word vector semantic analysis, including a processor and a memory; wherein, the memory stores a computer program, and the computer program is used to be loaded by the processor and execute the above-mentioned companion robot reliability evaluation method based on word vector semantic analysis.

[0136] like Figure 2 As shown, based on the same inventive concept as the above embodiment, this embodiment also provides a computer-readable storage medium, which stores instructions, and the instructions are used by the processor to load and execute the above-mentioned companion robot reliability evaluation method based on word vector semantic analysis.

[0137] In the embodiments of the mobile terminal and computer-readable storage medium provided in this application, all technical features of the above-mentioned control method embodiments are included. The expanded and explained contents of the specification are basically the same as those of the above-mentioned method embodiments and will not be repeated here.

[0138] An embodiment of the present application further provides a computer program product, which includes computer program code. When the computer program code runs on a computer, the computer executes the methods in the various possible implementation modes described above.

[0139] An embodiment of the present application also provides a chip, including a memory and a processor, wherein the memory is used to store computer programs, and the processor is used to call and run the computer programs from the memory, so that a device equipped with the chip executes the methods in the various possible implementation modes as described above.

[0140] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0141] In this application, the same or similar terminology, technical solutions and / or application scenario descriptions are generally only described in detail the first time they appear. When they appear again later, they are generally not repeated for the sake of brevity. When understanding the technical solutions and other contents of this application, for the same or similar terminology, technical solutions and / or application scenario descriptions that are not described in detail later, you can refer to the previous relevant detailed descriptions.

[0142] In this application, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0143] The various technical features of the technical solution of this application can be combined arbitrarily. In order to make the description concise, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0144] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product is stored in a storage medium as above, including a number of instructions for enabling a terminal device to execute the method of each embodiment of the present application. The above is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the description and drawings of this application, or directly or indirectly used in other related technical fields, is similarly included in the patent protection scope of the present application.

[0145] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referenced to each other.

[0146] The foregoing description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be readily conceived by a person skilled in the art within the technical scope disclosed herein are intended to be encompassed within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A reliability evaluation method for a care robot based on word vector semantic analysis, characterized in that: include: Obtaining the caregiver's input text data and the care robot's response text data; Encode the input text data and the response text data using a pre-trained word vector model to generate an input word vector sequence and a response word vector sequence, and determine a semantic difference feature vector between the input word vector sequence and the response word vector sequence; Determining a living habit difference coefficient and a language habit difference coefficient between the accompanying person and the accompanying robot, and determining a reliability correction coefficient of the accompanying robot based on the semantic difference feature vector, the living habit difference coefficient, and the language habit difference coefficient; Based on the reliability correction coefficient and the preset benchmark reliability index, a comprehensive reliability evaluation result of the accompanying robot is generated.

2. The reliability evaluation method of the accompanying robot based on word vector semantic analysis according to claim 1 is characterized in that: The word vector model is a BERT model; the use of the pre-trained word vector model to encode the input text data and the response text data to generate an input word vector sequence and a response word vector sequence specifically includes: Merging the input text data and the response text data through the classification identifier and the separation identifier to generate a joint input sequence; Processing the combined input sequence through the embedding layer of the BERT model to generate a word vector matrix; each row in the word vector matrix corresponds to a context-dependent vector of a subword; The word vector matrix boundary is positioned based on the positions of the classification identifier and the separation identifier, thereby separating the input word vector sequence and the response word vector sequence.

3. The reliability evaluation method of the accompanying robot based on word vector semantic analysis according to claim 2 is characterized in that: The merging and processing of the input text data and the response text data to generate a joint input sequence specifically includes: Adding a classification identifier to the headers of the input text data and the response text data; Adding a first separator between the input text data and the response text data; Adding a second separator mark at the end of the input text data and the response text data; Based on the classification identifier, the first separation identifier, and the second separation identifier, the input text data and the response text data are merged to form a joint input sequence.

4. The reliability evaluation method of the accompanying robot based on word vector semantic analysis according to claim 3 is characterized in that: Positioning the boundaries of the word vector matrix based on the positions of the classification identifier and the separation identifier, thereby separating the input word vector sequence and the response word vector sequence, specifically includes: Separate the input word vector sequence from the index position after the classification identifier to the index position before the first separation identifier in the index position of the joint input sequence; In the index positions of the joint input sequence, the response word vector sequence is separated from the index position after the first separation identifier to the index position before the second separation identifier.

5. The reliability evaluation method of the accompanying robot based on word vector semantic analysis according to claim 1 is characterized in that: The determining of the semantic difference feature vectors of the input word vector sequence and the response word vector sequence specifically includes: The input word vector sequence and the response word vector sequence are used as input to generate a bidirectional context feature vector through a Transformer model; the Transformer model includes a query space, a key space, and a value space, and the query space, the key space, and the value space correspond to a query weight matrix, a key weight matrix, and a value weight matrix, respectively; Taking the global semantic vector corresponding to the response word vector sequence in the bidirectional context feature vector to construct a query vector, taking the local feature vector corresponding to the input word vector sequence in the bidirectional context feature vector to construct a key vector, and generating a difference weight distribution by element-by-element subtraction of the query vector and the key vector; the query vector is obtained by multiplying the global semantic vector and the query weight matrix; the key vector is obtained by multiplying the local feature vector and the key weight matrix; The semantic difference feature vector is generated based on the difference weight distribution and weighted summation of the value vector; the value vector is obtained by multiplying the input word vector sequence by the value weight matrix.

6. The reliability evaluation method of the accompanying robot based on word vector semantic analysis according to claim 1 is characterized in that: The semantic difference feature vector is calculated and determined by the following formula: in, is the semantic difference feature vector; Indicates the The query vector of the sequence; Indicates the The key vector of the sequence; is the sequence length of the bidirectional context feature vector; Indicates the The dimension of the key vector of the sequence; is a value vector.

7. The reliability evaluation method of the accompanying robot based on word vector semantic analysis according to claim 1 is characterized in that: The determining of the difference coefficients of living habits and language habits between the accompanying person and the accompanying robot specifically includes: Obtaining the living habit difference coefficient based on historical interaction data between the caregiver and the care robot through a hierarchical analysis method; The similarity between the user's pronunciation data and the preset dialect database is calculated to obtain the language habit difference coefficient.

8. The reliability evaluation method of the accompanying robot based on word vector semantic analysis according to claim 7 is characterized in that: The method of obtaining the living habit difference coefficient based on the historical interaction data between the caregiver and the care robot by using the hierarchical analysis method specifically includes: Collecting historical interaction data between the caregiver and the care robot; the historical interaction data includes the caregiver's living habit label and the robot's response accuracy; Constructing a life habit evaluation index system based on the historical interaction data; the life habit evaluation index system includes multiple first-level indicators, each of which has corresponding second-level sub-indicators; The weight vector of the first-level indicator is calculated by the hierarchical analysis method, and the entropy weight method weight of the second-level sub-indicator is combined to generate the living habit difference coefficient.

9. The reliability evaluation method of the accompanying robot based on word vector semantic analysis according to claim 1 is characterized in that: After generating a comprehensive reliability evaluation result of the accompanying robot based on the reliability correction coefficient and the preset benchmark reliability index, the method further includes: Determining a confidence score for the comprehensive reliability evaluation result; If the confidence score is less than a preset confidence threshold, the accompanying robot is controlled to enter an adaptive learning mode; in the adaptive learning mode, the accompanying robot can optimize the word vector model according to the input text data obtained in real time.

10. A reliability evaluation system for accompanying robots based on word vector semantic analysis, characterized in that: It includes a processor and a memory; wherein the memory stores a computer program, and the computer program is used to be loaded by the processor and execute the accompanying robot reliability evaluation method based on word vector semantic analysis as described in any one of claims 1 to 9.

Images