Human-computer interaction method and device based on AI intelligent pen container

Abstract

The invention provides a man-machine interaction method and device based on an AI intelligent pen container, and is suitable for the technical field of data processing, and the method comprises the steps: processing historical dialogue information and question voice information, and according to a generated historical dialogue text vector, a question voice text vector, an adjacent region radius and adjacent region vector quantity information, obtaining a question voice text vector; generating an initial historical dialogue keyword vector and an initial question voice keyword vector; and generating a target historical dialogue keyword vector and a target question voice keyword vector according to the initial historical dialogue keyword vector, the initial question voice keyword vector and the feature mapping vector, and generating reply text information and reply voice information in combination with the historical dialogue text vector. According to the method, the keywords of the historical dialogue content and the current question content are accurately extracted, the specific context and the subdivided topic are clarified through the keywords, the reply information is generated by introducing the historical dialogue content reference, and the context understanding ability of the AI intelligent pen container is improved through deep semantic analysis.

Description

Technical Field

[0001] The present application belongs to the field of data processing technology, and in particular, relates to a human-computer interaction method and device based on an AI smart pen holder. Background Art

[0002] With the rapid development of artificial intelligence technology and people's growing demand for convenient interaction, AI smart pen holders have come into being to provide users with smarter and more efficient ways of obtaining and interacting with information. They are especially suitable for office and learning scenarios, helping people to record, query and process information more easily.

[0003] Existing AI pen holders usually integrate technologies such as speech recognition and natural language processing. Through speech recognition, the user's instant voice command is converted into text, and natural language processing technology is used to understand the semantics of the voice command, and then accurate answers are provided based on the semantic analysis results.

[0004] However, in the existing technology, the AI ​​dialogue process lacks in-depth mining of the dialogue content, and has difficulty capturing the semantic associations between keywords. It also lacks understanding of context and semantics, resulting in low accuracy in keyword extraction and low targeted response content. Summary of the invention

[0005] In view of this, the embodiments of the present application provide a human-computer interaction method and device based on an AI smart pen holder, which aims to solve the problems existing in the existing natural language processing technology, that is, the inability to accurately extract keywords from the conversation content provided by the user, and the large error in the context understanding of the conversation content, resulting in poor quality of the generated reply content.

[0006] A first aspect of an embodiment of the present application provides a human-computer interaction method based on an AI smart pen holder, comprising: Obtain historical conversation information and voice information of questions input by users; Converting the historical conversation information and the question voice information to generate a plurality of historical conversation text vectors and a plurality of question voice text vectors; Generate an initial historical conversation keyword vector and an initial question voice keyword vector according to the plurality of historical conversation text vectors, the plurality of question voice text vectors, a preset adjacent area radius, and preset adjacent area vector quantity information; Generate a target historical conversation keyword vector and a target question voice keyword vector according to the initial historical conversation keyword vector, the initial question voice keyword vector and a plurality of preset feature mapping vectors; Constructing a conversation text feature structure diagram according to the historical conversation text vector and the target historical conversation keyword vector; According to the dialogue text feature structure diagram and the target question voice text keyword vector, a reply text message and a reply voice message are generated.

[0007] A second aspect of the embodiment of the present application provides a human-computer interaction device based on an AI smart pen holder, comprising: The information acquisition module is used to obtain historical conversation information and voice information of questions input by users; A text vector generation module, used to convert the historical conversation information and question voice information to generate multiple historical conversation text vectors and multiple question voice text vectors; An initial keyword vector generation module, used to generate an initial historical conversation keyword vector and an initial question voice keyword vector according to the plurality of historical conversation text vectors, the plurality of question voice text vectors, a preset adjacent area radius, and preset adjacent area vector quantity information; A target keyword vector generation module, used to generate a target historical conversation keyword vector and a target question voice keyword vector according to the initial historical conversation keyword vector, the initial question voice keyword vector and a plurality of preset feature mapping vectors; A dialogue text feature structure graph construction module, used to construct a dialogue text feature structure graph according to the historical dialogue text vector and the target historical dialogue keyword vector; The reply information generation module is used to generate reply text information and reply voice information according to the dialogue text feature structure diagram and the target question voice text keyword vector.

[0008] Compared with the prior art, the embodiments of the present application have the following beneficial effects: preliminary extraction of keyword information is performed based on the semantic feature density in the historical conversation information and the question content information, and then the preliminary extracted keyword information is enhanced through a preset feature mapping vector, thereby filtering out redundant information mixed in during the preliminary extraction to achieve accurate extraction of keywords, and by introducing historical conversation information for semantic analysis, the semantic association features between each keyword in the historical conversation information and the question content information are captured, thereby improving the AI ​​smart pen holder's ability to understand the context of the conversation content and improving the accuracy and pertinence of the automatically generated reply information. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0010] Figure 1 This is a schematic diagram of the implementation process of the human-computer interaction method based on the AI ​​smart pen holder provided in the first embodiment of the present application; Figure 2 This is a schematic diagram of the implementation flow of the human-computer interaction method based on the AI ​​smart pen holder provided in the second embodiment of the present application; Figure 3 This is a schematic diagram of the implementation flow of the human-computer interaction method based on the AI ​​smart pen holder provided in Example 3 of the present application; Figure 4 This is a schematic diagram of the implementation flow of the human-computer interaction method based on the AI ​​smart pen holder provided in the fourth embodiment of the present application; Figure 5 This is a schematic diagram of the implementation flow of the human-computer interaction method based on the AI ​​smart pen holder provided in Example 5 of the present application; Figure 6 This is a schematic diagram of the implementation flow of the human-computer interaction method based on the AI ​​smart pen holder provided in Example 6 of the present application; Figure 7 This is a schematic diagram of the implementation flow of the human-computer interaction method based on the AI ​​smart pen holder provided in Example 7 of the present application; Figure 8 This is a schematic diagram of the implementation flow of the human-computer interaction method based on the AI ​​smart pen holder provided in Example 7 of the present application; Fig. 9 This is a schematic diagram of the implementation flow of the human-computer interaction method based on the AI ​​smart pen holder provided in Example 7 of the present application; Fig.10 It is a structural schematic diagram of a human-computer interaction device based on an AI smart pen holder provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0012] In order to illustrate the technical solution described in this application, a specific embodiment is provided below for illustration.

[0013] Figure 1 The following is a flowchart of the implementation of the human-computer interaction method based on the AI ​​smart pen holder provided in the first embodiment of the present application, which is described in detail as follows: Step S101, obtaining historical conversation information and voice information of questions input by the user.

[0014] In this embodiment, the historical conversation information can be extracted from the memory of the AI ​​smart pen holder, and can be the instruction information or question information input by the user in the past when using the AI ​​smart pen holder, as well as the reply content automatically generated by the AI ​​smart pen holder when used in the past. The question voice information can refer to the question or instruction information raised by the user to the AI ​​smart pen holder, which can be obtained through a component with a sound pickup function.

[0015] Step S102, converting the historical conversation information and question voice information to generate a plurality of historical conversation text vectors and a plurality of question voice text vectors.

[0016] In this embodiment, the historical conversation information may be preprocessed first, which may include operations such as removing special characters, punctuation marks, converting letters to upper and lower case letters, etc., to unify the text format, and then perform word segmentation processing, that is, splitting the sentence into multiple independent words or phrases, and then converting each word or phrase into a vector through a word embedding model to obtain a historical conversation text vector. The question voice text vector may be first converted into text content through speech recognition technology, and then the text content is preprocessed, word segmented, word embedded, and vector aggregated to generate multiple question voice text vectors.

[0017] Step S103, generating an initial historical conversation keyword vector and an initial question voice keyword vector according to the plurality of historical conversation text vectors, the plurality of question voice text vectors, a preset adjacent area radius, and preset adjacent area vector quantity information.

[0018] In this embodiment, the preset adjacent area radius and the preset adjacent area vector quantity information can be manually set. It can be understood that when the semantic feature information is densely distributed in the multidimensional space, that is, the similarity between the semantic feature information is high, the preset adjacent area radius can take a relatively small value. On the contrary, when the semantic feature information is sparsely distributed in the multidimensional space, that is, the similarity between the semantic feature information is low, the preset adjacent area radius can take a relatively large value. Among them, the preset adjacent area radius can be 0.1 to 0.5, or 1 to 5. The preset adjacent area vector quantity information can be related to the dimensions of the historical dialogue text vector and the question voice text vector, and the value can be 300.

[0019] It can be that the similarity between each historical conversation text vector and each question voice text vector is calculated first, and then the text vector information classified into the adjacent area is determined according to the size relationship between the similarity and the preset adjacent area radius, and then the number of text vector information in each adjacent area is counted, and the size relationship between the number of text vector information in the adjacent area and the preset adjacent area vector quantity information is determined, and the adjacent area whose text vector information number is greater than the preset adjacent area vector quantity information is screened, and the text vector information in the screened adjacent area can be used as the initial historical conversation keyword vector and the initial question voice keyword vector; it can also be that the text vector information in the screened adjacent area is calculated for similarity, and the text vector information with high similarity is aggregated, and the aggregated vector is used as the initial historical conversation keyword vector and the initial question voice keyword vector. Among them, the similarity can be cosine similarity. This embodiment is used to perform preliminary extraction of historical conversation keywords and question voice text keywords based on the density of semantic features in historical conversation text information and question voice text information.

[0020] Step S104, generating a target historical conversation keyword vector and a target question voice keyword vector according to the initial historical conversation keyword vector, the initial question voice keyword vector and a plurality of preset feature mapping vectors.

[0021] In this embodiment, multiple preset feature mapping vectors can be manually set. It can be achieved by performing dot multiplication operations on multiple preset feature mapping vectors with the initial historical conversation keyword vector and the initial question voice keyword vector, thereby widening the distance between each value in each historical conversation keyword vector and question voice keyword vector, so as to achieve enhanced processing of the keyword vector initially extracted and avoid redundant information. The enhanced historical conversation keyword vector and question voice keyword vector are used as the target historical conversation keyword vector and target question voice keyword vector. It can be understood that in the existing natural language processing technology, it is often difficult to accurately capture semantic associations in keyword extraction, and it is easy to ignore the context, resulting in one-sided extracted keywords. This embodiment improves the accuracy of keyword extraction by deeply extracting and analyzing semantic features.

[0022] Step S105, constructing a dialogue text feature structure diagram according to the historical dialogue text vector and the target historical dialogue keyword vector.

[0023] In this embodiment, the historical conversation text vectors and the target historical conversation keyword vectors may be used as nodes of the conversation text feature structure graph, and the logical distances between the historical conversation text vectors and the distances between the target historical conversation keyword vectors may be calculated as edges between the nodes, thereby constructing the conversation text feature structure graph through the nodes and edges.

[0024] Step S106, generating reply text information and reply voice information according to the dialogue text feature structure diagram and the target question voice text keyword vector.

[0025] In this embodiment, the target question speech text keyword vector can be randomly inserted into the dialogue text feature structure diagram as a new node of the dialogue text feature structure diagram, and the original node of the dialogue text feature structure diagram is used as the old node, and then the logical distance between the new node and each old node is calculated. A logical distance threshold can be preset artificially. When the logical distance between the new node and each old node is less than the logical distance threshold, the old node is used as a reference node, and the logical distance between the reference node and the new node is used as a weight coefficient. The new node and the reference node are weighted and summed, and the results calculated for multiple new nodes are semantically spliced ​​as a reply text vector, and then the reply text vector is decoded by the language model to generate a reply text message, and then the reply text message is converted into a reply voice message by the speech synthesis technology. Among them, the reply text message can be visually output through the electronic screen of the AI ​​smart pen holder, and the reply voice message can be output through the speaker of the AI ​​smart pen holder. It is understandable that in the existing human-computer dialogue technology, usually only the question information currently provided by the user is analyzed and processed, which leads to large errors in the contextual understanding of the dialogue content, shallow analysis of complex semantics, and difficulty in accurately grasping the user's true intentions, resulting in poor quality of the generated reply content. This embodiment introduces historical dialogue content for reference and conducts deep semantic analysis to improve the AI ​​smart pen holder's ability to understand the context.

[0026] The human-computer interaction method based on the AI ​​smart pen holder provided in the embodiment of the present application performs preliminary extraction of keyword information based on the semantic feature density in the historical conversation information and the question content information, and then enhances the preliminary extracted keyword information through a preset feature mapping vector, thereby filtering out redundant information mixed in during the preliminary extraction to achieve accurate extraction of keywords, and by introducing historical conversation information for semantic analysis, the semantic association features between each keyword in the historical conversation information and the question content information are captured, thereby improving the AI ​​smart pen holder's ability to understand the context of the conversation content and improving the accuracy and pertinence of the automatically generated reply information.

[0027] Figure 2The flowchart of the implementation of the human-computer interaction method based on the AI ​​smart pen holder provided in the second embodiment of the present application is shown. The difference between the second embodiment and the first embodiment is that the step S103 specifically includes: Step S201, generating a human-computer interaction text vector set according to a plurality of the historical conversation text vectors and a plurality of the question voice text vectors; the human-computer interaction text vector set includes a plurality of human-computer interaction text vectors.

[0028] In this embodiment, multiple historical conversation text vectors and multiple question voice text vectors may be concatenated to generate a human-computer interaction text vector set, which is used for subsequent preliminary semantic extraction of keyword features in the historical conversation text and question voice text. It can be understood that the elements in the human-computer interaction text vector set are used as human-computer interaction text vectors to determine the density of semantic feature information in multidimensional space in the subsequent calculation process.

[0029] Step S202: randomly extracting elements of the human-computer interaction text vector set to obtain a human-computer interaction text center vector.

[0030] In this embodiment, an element may be randomly extracted from the human-computer interaction text vector set, that is, a human-computer interaction text vector may be randomly extracted as the human-computer interaction text center vector.

[0031] Step S203, calculating the Euclidean distance between each of the human-computer interaction text vectors and the human-computer interaction text center vector to obtain the interaction text vector distance.

[0032] In this embodiment, the Euclidean distance between each human-computer interaction text vector and the human-computer interaction text center vector may be calculated first to obtain the interactive text vector distances of multiple human-computer interaction text vectors corresponding to the human-computer interaction text center vector.

[0033] Step S204, determine whether the interactive text vector distance is less than or equal to a preset adjacent area radius; if so, use the human-computer interaction text vector corresponding to the interactive text vector distance as the human-computer interaction text adjacent area vector; if not, skip processing the human-computer interaction text vector corresponding to the interactive text vector distance.

[0034] In this embodiment, the preset adjacent area radius can be set manually. When the interactive text vector distance is less than or equal to the preset adjacent area radius, it means that the distance between the human-computer interaction text vector and the human-computer interaction text center vector in the vector space is small, that is, the semantic feature similarity is high, and the human-computer interaction text vector can be classified into the adjacent area of ​​the human-computer interaction text center vector, as a semantic feature representing a keyword in the subsequent synthesis of the keyword vector; when the interactive text vector distance is greater than the preset adjacent area radius, it means that the distance between the human-computer interaction text vector and the human-computer interaction text center vector in the vector space is large, that is, the semantic feature similarity is low, and the human-computer interaction text vector cannot be classified into the adjacent area of ​​the human-computer interaction text center vector, then the human-computer interaction text vector corresponding to the interactive text vector distance is skipped, and the current human-computer interaction text vector is still present as a human-computer interaction text vector in the human-computer interaction text vector set, and then it is determined whether the distance between the next human-computer interaction text vector and the human-computer interaction text center vector can classify the next human-computer interaction text vector into the adjacent area of ​​the human-computer interaction text center vector. Until all human-computer interaction text vectors are traversed, the adjacent area of ​​the human-computer interaction text center vector is generated.

[0035] Step S205, generating an initial historical dialogue keyword vector and an initial question voice keyword vector according to the human-computer interaction text adjacent region vector and preset adjacent region vector quantity information.

[0036] In this embodiment, the number of human-computer interaction text adjacent area vectors of the adjacent areas of each human-computer interaction text center vector can be first counted, and the adjacent areas whose number of human-computer interaction text adjacent area vectors exceeds the preset adjacent area vector number information are used as a set for generating keyword vectors, and the vectors in the set are extracted according to the source, the elements derived from the historical dialogue information are extracted to generate the initial historical dialogue keyword vector, and the elements derived from the question voice information are extracted to generate the initial question voice keyword vector.

[0037] The human-computer interaction method based on the AI ​​smart pen holder provided in the embodiment of the present application automatically identifies complex semantic features with irregular logical shapes in the vector space, and uses the automatic identification results as adjacent areas to realize flexible category division of semantic features. At the same time, the semantic feature vectors that cannot be divided into adjacent areas are filtered out, thereby improving the accuracy of keyword extraction of the AI ​​smart pen holder, thereby improving the quality and efficiency of human-computer interaction.

[0038] Figure 3 The flowchart of the implementation of the human-computer interaction method based on the AI ​​smart pen holder provided in the third embodiment of the present application is shown. The difference between the third embodiment and the second embodiment is that the step S205 specifically includes: Step S301, counting the number of adjacent region vectors of the human-computer interaction text to obtain information on the number of neighborhood vectors of the interaction text.

[0039] In this embodiment, the number of adjacent area vectors of all human-computer interaction texts is counted, that is, the number of adjacent area vectors of human-computer interaction texts adjacent to the center vector of each human-computer interaction text is counted to obtain the information on the number of interactive text neighborhood vectors, which is used to subsequently determine whether the adjacent area is valid.

[0040] Step S302, determine whether the interactive text neighborhood vector quantity information is greater than or equal to the preset adjacent area vector quantity information; if so, proceed to step S303; if not, proceed to step S304.

[0041] In this embodiment, when the interactive text neighborhood vector quantity information is greater than or equal to the preset adjacent area vector quantity information, it means that the density of semantic feature points in the adjacent area is large, and the adjacent area is highly effective and can be used to generate keyword information; when the interactive text neighborhood vector quantity information is less than the preset adjacent area vector quantity information, it means that the density of semantic feature points in the adjacent area is small, and the adjacent area is low in effectiveness, and whether it can be used to generate keyword information requires further judgment later.

[0042] Step S303: determining the human-computer interaction text center vector as the human-computer interaction text core vector.

[0043] In this embodiment, when the number of interactive text neighborhood vectors is greater than or equal to the preset number of adjacent area vectors, it means that the density of semantic feature points in the adjacent area is large, and the adjacent area is highly effective and can be used to generate keyword information. All vectors in the adjacent area can be determined as core vectors of human-computer interactive text.

[0044] Step S304: determining the central vector of the human-computer interaction text as a non-core vector of the human-computer interaction text.

[0045] In this embodiment, when the number of interactive text neighborhood vectors is less than the preset adjacent area vector number information, it means that the density of semantic feature points in the adjacent area is small, and the effectiveness of the adjacent area is low. Whether it can be used to generate keyword information requires further judgment. In this case, all vectors in the adjacent area can be determined as non-core vectors of human-computer interactive text.

[0046] Step S305 , counting the number of the human-computer interaction text core vectors and the number of the human-computer interaction text non-core vectors to obtain the number of interaction text core vectors and the number of interaction text non-core vectors.

[0047] In this embodiment, the number of all human-computer interaction text core vectors is counted, and the statistical result is used as the number of interaction text core vectors; the number of all human-computer interaction text non-core vectors is counted, and the statistical result is used as the number of interaction text non-core vectors.

[0048] Step S306, determining whether the sum of the number of interactive text core vectors and the number of interactive text non-core vectors is equal to the number of elements in the human-computer interactive text vector set; if so, proceeding to step S307; if not, proceeding to step S308.

[0049] In this embodiment, it is used to determine whether all human-computer interaction text vectors in the human-computer interaction text vector set have been traversed; when the sum of the number of interactive text core vectors and the number of interactive text non-core vectors is equal to the number of elements in the human-computer interaction text vector set, it means that all human-computer interaction text vectors in the human-computer interaction text vector set have been traversed, and all human-computer interaction text vectors have been updated to human-computer interaction text core vectors or human-computer interaction text non-core vectors, and no traversal calculation is required, and keyword vector synthesis processing can be performed through the human-computer interaction text core vector and the human-computer interaction text non-core vector; when the sum of the number of interactive text core vectors and the number of interactive text non-core vectors is less than the number of elements in the human-computer interaction text vector set, it means that all human-computer interaction text vectors in the human-computer interaction text vector set have not been traversed, and it is necessary to re-iterate the calculation to ensure that all human-computer interaction text vectors are traversed and all human-computer interaction text vectors are updated to human-computer interaction text core vectors or human-computer interaction text non-core vectors.

[0050] Step S307: Generate an initial history dialogue keyword vector and an initial question voice keyword vector according to the human-computer interaction text core vector and the human-computer interaction text non-core vector.

[0051] In this embodiment, when the sum of the number of interactive text core vectors and the number of interactive text non-core vectors is equal to the number of elements in the human-computer interaction text vector set, it means that all human-computer interaction text vectors in the human-computer interaction text vector set have been traversed, and all human-computer interaction text vectors have been updated to human-computer interaction text core vectors or human-computer interaction text non-core vectors. There is no need to perform traversal calculations, and keyword vector synthesis processing can be performed through the human-computer interaction text core vectors and human-computer interaction text non-core vectors.

[0052] Step S308, obtaining a remaining interaction text vector set according to the human-computer interaction text vector set, the human-computer interaction text core vector and the human-computer interaction text non-core vector.

[0053] In this embodiment, when the sum of the number of interactive text core vectors and the number of interactive text non-core vectors is less than the number of elements in the human-computer interactive text vector set, it means that all human-computer interactive text vectors in the human-computer interactive text vector set have not been traversed, and it is necessary to re-iterate and calculate, and it is necessary to determine the human-computer interactive text vectors that have not been traversed. The elements in the human-computer interactive text vector set that have been updated to human-computer interactive text core vectors or human-computer interactive text non-core vectors can be eliminated to generate a remaining interactive text vector set for performing the traversal operation of the remaining human-computer interactive text vectors.

[0054] Step S309: taking the remaining interaction text vector set as the human-computer interaction text vector set, and returning to step S202.

[0055] In this embodiment, the remaining interactive text vector set is used as the human-computer interaction text vector set, and the elements are randomly re-extracted, the similarity is calculated, and the adjacent areas are divided, so as to realize recursive operation, which is used to ensure that all human-computer interaction text vectors are updated to human-computer interaction text core vectors or human-computer interaction text non-core vectors.

[0056] The human-computer interaction method based on the AI ​​smart pen holder provided in the embodiment of the present application avoids the omission of key semantic features in historical dialogue information and question content information by traversing and updating all human-computer interaction text vectors, and continuously expands the number of semantic feature vectors that can be used to generate keywords by continuously incorporating human-computer interaction text vectors into adjacent areas, thereby ensuring the accuracy and comprehensiveness of the generated keywords, thereby enabling the AI ​​smart pen holder to understand the dialogue topic more accurately, and then provide reply information that is more in line with the context, so as to effectively improve the human-computer interaction experience of the AI ​​smart pen holder.

[0057] Figure 4 The flowchart of the implementation of the human-computer interaction method based on the AI ​​smart pen holder provided in the fourth embodiment of the present application is shown. The difference between the fourth embodiment and the third embodiment is that the step S307 specifically includes: Step S401, calculating the Euclidean distance between each non-core vector of the human-computer interaction text and each core vector of the human-computer interaction text to obtain the confirmation distance of the non-core point of the interaction text.

[0058] In this embodiment, the Euclidean distance between each non-core vector of the human-computer interaction text and each core vector of the human-computer interaction text is calculated, and the calculation result is used as the confirmation distance of the non-core point of the interaction text to further determine the validity of each non-core vector of the human-computer interaction text.

[0059] Step S402, determining whether the confirmation distance of the non-core point of the interactive text is less than or equal to a preset adjacent area radius; if so, proceeding to step S403; if not, proceeding to step S404.

[0060] In this embodiment, when the confirmation distance of the non-core point of the interactive text is less than or equal to the preset adjacent area radius, it means that the semantic feature information represented by the non-core vector of the human-computer interactive text has a strong correlation with other semantic feature information, and the semantic validity of the non-core vector of the human-computer interactive text is high, and it can be used to generate keyword information. When the confirmation distance of the non-core point of the interactive text is greater than the preset adjacent area radius, it means that the semantic feature information represented by the non-core vector of the human-computer interactive text has a weak correlation with other semantic feature information, and the semantic validity of the non-core vector of the human-computer interactive text is low, and it can be used to generate keyword information.

[0061] Step S403: taking the human-computer interaction text non-core vector corresponding to the interaction text non-core point confirmation distance as the human-computer interaction text edge point vector.

[0062] In this embodiment, when the confirmation distance of the non-core point of the interactive text is less than or equal to the preset adjacent area radius, it means that the semantic feature information represented by the non-core vector of the human-computer interaction text is strongly correlated with other semantic feature information, and the semantic validity of the non-core vector of the human-computer interaction text is high. The non-core vector of the human-computer interaction text can be used as an edge point vector of the human-computer interaction text to generate keyword information.

[0063] Step S404: taking the human-computer interaction text non-core vector corresponding to the interaction text non-core point confirmation distance as the human-computer interaction text redundant point vector.

[0064] In this embodiment, when the confirmation distance of the non-core point of the interactive text is greater than the preset adjacent area radius, it means that the semantic feature information represented by the non-core vector of the human-computer interaction text has a weak correlation with other semantic feature information, and the semantic validity of the non-core vector of the human-computer interaction text is low. The non-core vector of the human-computer interaction text is then treated as a redundant point vector of the human-computer interaction text, that is, it is eliminated as a noise component and is not used to generate keyword information.

[0065] Step S405 , generating a plurality of interaction text keyword vectors according to the human-computer interaction text core vector and the human-computer interaction text edge point vector.

[0066] In this embodiment, the semantic content represented by the core vector of the human-computer interaction text can be used as the main reference information, and the semantic content represented by the edge point vector of the human-computer interaction text can be used as the auxiliary reference information. The interactive text keyword vector includes a main keyword vector and an auxiliary keyword vector. The core vector of the human-computer interaction text is aggregated to generate a main keyword vector; the edge point vector of the human-computer interaction text is aggregated to generate an auxiliary keyword vector. In the subsequent calculation, a higher weight can be assigned to the main keyword vector, and a lower weight can be assigned to the auxiliary keyword vector.

[0067] Step S406, performing numerical separation and extraction processing on the interactive text keyword vector to generate an initial historical dialogue keyword vector and an initial question voice keyword vector.

[0068] In this embodiment, the semantic features represented in the interactive text keyword vector may be separated according to different sources to obtain the initial historical conversation keyword vector and the initial question voice keyword vector.

[0069] The human-computer interaction method based on the AI ​​smart pen holder provided in the embodiment of the present application determines the key semantic feature information in the conversation content through the core vector of the human-computer interaction text, which is used to quickly lock the core content in the conversation information, improve the accuracy and effectiveness of the AI ​​smart pen holder in grasping the main idea of ​​the conversation content, expand the semantic feature information in the conversation content through the edge point vector of the human-computer interaction text, incorporate the conversation content closely related to the core content, enrich the content of keywords, and improve the comprehensiveness of keyword extraction, thereby helping the AI ​​smart pen holder to provide users with more detailed reply information that is more in line with user needs, and effectively improve the quality and efficiency of human-computer interaction.

[0070] Figure 5 The flowchart of the implementation of the human-computer interaction method based on the AI ​​smart pen holder provided in the fifth embodiment of the present application is shown, which differs from the above-mentioned first embodiment in that: The plurality of preset feature mapping vectors include a preset search mapping vector, a preset index mapping vector and a preset content mapping vector; The step S104 specifically includes: Step S501, based on the initial historical conversation keyword vector, the initial question voice keyword vector, the preset retrieval mapping vector, the preset index mapping vector and the preset content mapping vector, calculate the initial historical conversation keyword retrieval vector, the initial historical conversation keyword index vector, the initial historical conversation keyword content vector, the initial question voice keyword retrieval vector, the initial question voice keyword index vector and the initial question voice keyword content vector.

[0071] In this embodiment, the preset retrieval mapping vector, the preset index mapping vector, and the preset content mapping vector can all be manually set to further distance the semantic features in the vector space. The initial historical conversation keyword vector and the preset retrieval mapping vector may be subjected to a dot product operation to obtain the initial historical conversation keyword retrieval vector; the initial historical conversation keyword vector and the preset index mapping vector may be subjected to a dot product operation to obtain the initial historical conversation keyword index vector; the initial historical conversation keyword vector and the preset content mapping vector may be subjected to a dot product operation to obtain the initial historical conversation keyword content vector; the initial question voice keyword vector and the preset retrieval mapping vector may be subjected to a dot product operation to obtain the initial question voice keyword retrieval vector; the initial question voice keyword vector and the preset index mapping vector may be subjected to a dot product operation to obtain the initial question voice keyword index vector; the initial question voice keyword vector and the preset content mapping vector may be subjected to a dot product operation to obtain the initial question voice keyword content vector.

[0072] Step S502: Calculate the initial historical conversation keyword query variable according to the initial historical conversation keyword search vector and the initial historical conversation keyword index vector.

[0073] In this embodiment, a dot product operation may be performed on the initial historical conversation keyword search vector and the transposed matrix of the initial historical conversation keyword index vector, and the calculation result is used as the initial historical conversation keyword query variable.

[0074] Step S503: Calculate the initial question voice keyword query variable according to the initial question voice keyword search vector and the initial question voice keyword index vector.

[0075] In this embodiment, a dot product operation may be performed on the initial question voice keyword search vector and the transposed matrix of the initial question voice keyword index vector, and the calculation result is used as the initial question voice keyword query variable.

[0076] Step S504, normalizing the initial historical conversation keyword query variable and the initial question voice keyword query variable to obtain an initial historical keyword query scaling variable and an initial question keyword query scaling variable.

[0077] In this embodiment, the normalization process can be implemented by a normalization function, a Softmax function, or a Sparsemax function. The initial historical conversation keyword query variable and the initial question voice keyword query variable can be used as independent variables of the normalization function, and the calculated function values ​​can be used as the initial historical keyword query scaling variable and the initial question keyword query scaling variable.

[0078] Step S505 , calculating a target historical conversation keyword vector according to the initial historical keyword query scaling variable and the initial historical conversation keyword content vector.

[0079] In this embodiment, the initial historical keyword query scaling variable and the initial historical conversation keyword content vector may be multiplied, and the multiplication result is used as the target historical conversation keyword vector.

[0080] Step S506, calculating a target question voice keyword vector according to the initial question keyword query scaling variable and the initial question voice keyword content vector.

[0081] In this embodiment, the initial question keyword query scaling variable and the initial question voice keyword content vector may be multiplied, and the multiplication result is used as the target question voice keyword vector.

[0082] The human-computer interaction method based on the AI ​​smart pen holder provided in the embodiment of the present application captures semantic feature information in different logical subspaces through multiple preset feature mapping vectors, so that multiple fields and levels involved in the conversation content can be fully analyzed, thereby extracting keywords for different fields or levels, and accurately identifying the user's historical conversation content and current question content in specific fields such as knowledge questions and answers or emotional communication, thereby providing more refined keywords and improving the pertinence and effectiveness of generated reply information.

[0083] Figure 6 The flowchart of the implementation of the human-computer interaction method based on the AI ​​smart pen holder provided in the sixth embodiment of the present application is shown. The difference between the sixth embodiment and the first embodiment is that the step S105 specifically includes: Step S601: taking the target historical conversation keyword vector as the main node.

[0084] In this embodiment, the keyword content corresponding to the target historical conversation keyword vector is used as the main reference information, and the target historical conversation keyword vector is used as the main node of the conversation text feature structure graph. It can be understood that the conversation text feature structure graph is constructed based on the vector space.

[0085] Step S602, obtaining a historical conversation auxiliary text vector according to the historical conversation text vector and the target historical conversation keyword vector.

[0086] In this embodiment, the content in the historical conversation information except the content represented by the target historical conversation keyword vector is used as auxiliary reference information, and the auxiliary reference information in the historical conversation information is converted into a vector through word embedding technology as the historical conversation auxiliary text vector.

[0087] Step S603: Use the historical conversation auxiliary text vector as an auxiliary node.

[0088] In this embodiment, the content in the historical conversation information except the content represented by the target historical conversation keyword vector is used as auxiliary reference information, and the auxiliary reference information in the historical conversation information is converted into a historical conversation auxiliary text vector through word embedding technology, and the historical conversation auxiliary text vector is used as an auxiliary node of the conversation text feature structure diagram.

[0089] Step S604, calculating the similarity between each of the primary nodes and the similarity between each of the secondary nodes.

[0090] In this embodiment, the Euclidean distance between each main node and the Euclidean distance between each auxiliary node may be calculated, and the similarity between each main node and the similarity between each auxiliary node may be calculated based on the Euclidean distance to quantify the semantic relevance between each main node and each auxiliary node.

[0091] Step S605, calculating the similarity between each of the primary nodes and each of the secondary nodes to obtain the similarity between the primary and secondary nodes.

[0092] In this embodiment, the Euclidean distance between each main node and each auxiliary node may be calculated, and the similarity between each main node and each auxiliary node may be calculated based on the Euclidean distance to quantify the semantic relevance between each main and auxiliary node.

[0093] Step S606, determining whether the similarity between the master nodes is greater than a preset similarity threshold; if so, generating edges between the master nodes; if not, not generating edges between the master nodes.

[0094] In this embodiment, when the similarity between the main nodes is greater than the preset similarity threshold, it means that the semantic features represented by the two main nodes corresponding to the similarity are strongly correlated, and an edge between the two main nodes can be generated to characterize that the two main nodes have a strong correlation; when the similarity between the main nodes is less than or equal to the preset similarity threshold, it means that the semantic features represented by the two main nodes corresponding to the similarity are weakly correlated, and an edge between the two main nodes may not be generated to characterize that the two main nodes have a weak correlation or no correlation.

[0095] Step S607, determining whether the similarity between the auxiliary nodes is greater than a preset similarity threshold; if so, generating edges between the auxiliary nodes; if not, not generating edges between the auxiliary nodes.

[0096] In this embodiment, when the similarity between the auxiliary nodes is greater than the preset similarity threshold, it means that the semantic features represented by the two auxiliary nodes corresponding to the similarity are strongly correlated, and an edge between the two auxiliary nodes can be generated to characterize that the two auxiliary nodes have a strong correlation; when the similarity between the auxiliary nodes is less than or equal to the preset similarity threshold, it means that the semantic features represented by the two auxiliary nodes corresponding to the similarity are weakly correlated, and an edge between the two auxiliary nodes may not be generated to characterize that the two auxiliary nodes have a weak correlation or no correlation.

[0097] Step S608, determining whether the similarity between the primary and secondary nodes is greater than a preset similarity threshold; if so, generating an edge between the primary and secondary nodes; if not, not generating an edge between the primary and secondary nodes.

[0098] In this embodiment, when the similarity between the main and auxiliary nodes is greater than the preset similarity threshold, it means that the semantic features represented by the main node and the auxiliary node corresponding to the similarity are strongly correlated, and an edge between the main and auxiliary nodes can be generated to represent that the two nodes have a strong correlation; when the similarity between the main and auxiliary nodes is less than or equal to the preset similarity threshold, it means that the semantic features represented by the main node and the auxiliary node corresponding to the similarity are weakly correlated, and an edge between the two nodes may not be generated to represent that the two nodes have a weak correlation or no correlation.

[0099] Step S609, constructing a conversation text feature structure graph according to the main node, the auxiliary node, the edge between the main nodes, the edge between the auxiliary nodes, and the edge between the main and auxiliary nodes.

[0100] In this embodiment, the dialogue text feature structure diagram may be generated by combining the main nodes, the auxiliary nodes, the edges between the main nodes, the edges between the auxiliary nodes, and the edges between the main and auxiliary nodes in the vector space.

[0101] The human-computer interaction method based on the AI ​​smart pen holder provided in the embodiment of the present application models the complex relationship between the historical conversation content and the current question content by constructing a conversation text feature structure diagram in a multi-dimensional vector space, thereby introducing the historical conversation content for reference, and capturing the semantic correlation between the keywords in the conversation process, the historical conversation content, and the question content in the subsequent processing process, mining the potential semantic information in the historical conversation content and the question content, and combining the mining results in the historical conversation content with the current question content to generate reply information, so that the AI ​​intelligence can understand the coherence and pertinence in the conversation process, and enhance the AI ​​smart pen holder's ability to deeply understand the conversation context content, thereby providing users with a more comprehensive reply.

[0102] Figure 7The flowchart of the implementation of the human-computer interaction method based on the AI ​​smart pen holder provided in the seventh embodiment of the present application is shown. The difference between the seventh embodiment and the sixth embodiment is that the step S106 specifically includes: Step S701: taking the target question speech text keyword vector as a new node.

[0103] In this embodiment, the target question speech text keyword vector is used as a new node in the dialogue text feature structure graph.

[0104] Step S702: randomly insert the new node into the conversation text feature structure graph to obtain an updated conversation text feature structure graph.

[0105] In this embodiment, the target question speech text keyword vector is used as a new node in the dialogue text feature structure diagram and randomly inserted into the dialogue text feature structure diagram to establish a connection between the question content and the historical dialogue content, realize real-time information fusion, and enable the AI ​​smart pen holder to control the dialogue context of the current question content. The dialogue text feature structure diagram after inserting the new node is used as the updated dialogue text feature structure diagram.

[0106] Step S703, calculating the similarity between the new node and the main node to obtain the new main node similarity.

[0107] In this embodiment, the Euclidean distance between the new node and the main node may be calculated, and the calculated Euclidean distance is output as the new main node similarity.

[0108] Step S704, calculating the similarity between the new node and the auxiliary node to obtain the new auxiliary node similarity.

[0109] In this embodiment, the Euclidean distance between the new node and the auxiliary node may be calculated, and the calculated Euclidean distance is output as the new auxiliary node similarity.

[0110] Step S705, determining whether the new master node similarity is greater than a preset similarity threshold; if so, generating an edge between the new node and the master node; if not, not generating an edge between the new node and the master node.

[0111] In this embodiment, when the similarity of the new main node is greater than the preset similarity threshold, it means that the semantic features represented by the main node and the new node corresponding to the similarity are strongly correlated, and an edge between the new node and the main node can be generated to characterize that the two nodes have a strong correlation; when the similarity between the new main nodes is less than or equal to the preset similarity threshold, it means that the semantic features represented by the main node and the new node corresponding to the similarity are weakly correlated, and an edge between the two nodes may not be generated to characterize that the two nodes have a weak correlation or no correlation.

[0112] Step S706, determining whether the similarity of the new auxiliary node is greater than a preset similarity threshold; if so, generating an edge between the new node and the auxiliary node; if not, not generating an edge between the new node and the auxiliary node.

[0113] In this embodiment, when the similarity of the new auxiliary nodes is greater than the preset similarity threshold, it means that the semantic features represented by the auxiliary nodes corresponding to the similarity and the new node are strongly correlated, and an edge between the new node and the auxiliary node can be generated to characterize that the two nodes have a strong correlation; when the similarity between the new auxiliary nodes is less than or equal to the preset similarity threshold, it means that the semantic features represented by the auxiliary nodes corresponding to the similarity and the new node are weakly correlated, and an edge between the two nodes may not be generated to characterize that the two nodes have a weak correlation or no correlation.

[0114] Step S707, generating reply text information and reply voice information according to the new node, main node, auxiliary node, edge between main nodes, edge between auxiliary nodes, edge between main and auxiliary nodes, edge between new node and main node, edge between new node and auxiliary node.

[0115] In this embodiment, different weight coefficients may be set for the new node, the main node, and the auxiliary node first. The connection paths between the new node, the main node, and the auxiliary node in the updated dialogue text feature structure diagram may be determined based on the edges between the main nodes, the edges between the auxiliary nodes, the edges between the main and auxiliary nodes, the edges between the new node and the main node, and the edges between the new node and the auxiliary node. The weighted summation calculation is performed on the new nodes, the main nodes, and the auxiliary nodes on the multiple connection paths connected by the edges, and the calculated result is used as the reply text vector. The reply text vector is then decoded by the language model to generate a reply text message, and then the reply text message is converted into a reply voice message by the speech synthesis technology. Among them, the reply text message can be visualized and output through the electronic screen of the AI ​​smart pen holder, and the reply voice message can be output through the speaker of the AI ​​smart pen holder.

[0116] The human-computer interaction method based on the AI ​​smart pen holder provided in the embodiment of the present application can build a complete semantic association network. Since each question asked by a user is new information, after the new information is added as a new node, it can be connected with the historical conversation content through the semantic association network to form a continuously expanding conversation text feature structure diagram, so that the AI ​​smart pen holder can understand the conversation context more comprehensively. When facing questions from different users and different types, new nodes can continuously enrich the semantic association network, so that the AI ​​smart pen holder can better adapt to diverse conversation needs and provide users with a high-quality interactive experience.

[0117] Figure 8The flowchart of the implementation of the human-computer interaction method based on the AI ​​smart pen holder provided in the eighth embodiment of the present application is shown. The difference between the eighth embodiment and the seventh embodiment is that the step S707 specifically includes: Step S801, counting the number of new nodes, the number of primary nodes, and the number of secondary nodes.

[0118] In this embodiment, the number of new nodes, the number of main nodes, and the number of auxiliary nodes in the updated dialogue text feature structure graph are counted for subsequent determination of the dimension of the adjacency variable matrix.

[0119] Step S802: Calculate the total number of nodes according to the number of new nodes, the number of primary nodes, and the number of secondary nodes.

[0120] In this embodiment, the number of new nodes, the number of primary nodes, and the number of secondary nodes are summed, and the sum result is used as the total number of nodes.

[0121] Step S803, generating an adjacency variable matrix according to the total number of nodes, edges between primary nodes, edges between secondary nodes, edges between primary and secondary nodes, edges between new nodes and primary nodes, and edges between new nodes and secondary nodes.

[0122] In this embodiment, an empty matrix may be set first, and the total number of nodes is used as the dimension of the matrix, and then the entire updated dialogue text feature structure diagram is traversed. When the edges between the main nodes, the edges between the auxiliary nodes, the edges between the main and auxiliary nodes, the edges between the new node and the main node, and the edges between the new node and the auxiliary node are traversed, the elements corresponding to the corresponding rows and columns in the empty matrix are set to 1; if the edges between the main nodes, the edges between the auxiliary nodes, the edges between the main and auxiliary nodes, the edges between the new node and the main node, and the edges between the new node and the auxiliary node are not traversed, the elements corresponding to the corresponding rows and columns in the empty matrix are set to 0. The finally generated matrix is ​​output as an adjacency variable matrix.

[0123] Step S804, generating reply text information and reply voice information according to the adjacency variable matrix, the main node, the auxiliary node and the new node.

[0124] In this embodiment, the main node, auxiliary node, new node and node information matrix can be constructed, and the elements of the node information matrix are the main node, auxiliary node and new node. The node information matrix can be multiplied by the adjacent variable matrix to obtain a reply text vector, and then the reply text vector is decoded by the language model to generate a reply text message, and then the reply text message is converted into a reply voice message by speech synthesis technology. Among them, the reply text message can be visualized and output through the electronic screen of the AI ​​smart pen holder, and the reply voice message can be output through the speaker of the AI ​​smart pen holder.

[0125] The human-computer interaction method based on the AI ​​smart pen holder provided in the embodiment of the present application can clearly present the association structure between each information node in the conversation by constructing an adjacency variable matrix. The main node represents the key topic, the auxiliary node is the relevant details, and the new node is the current question. These associations can be accurately expressed in mathematical form through the adjacency variable matrix, allowing the AI ​​smart pen holder to understand the relationship between information, thereby understanding the connection between different user questions and topics in historical conversation information, and quickly locating relevant information in historical conversation information, so that the reply information generated by the AI ​​smart pen holder is closer to the context, so as to enhance the user experience.

[0126] Fig. 9 The flowchart of the implementation of the human-computer interaction method based on the AI ​​smart pen holder provided in the ninth embodiment of the present application is shown. The difference between the ninth embodiment and the eighth embodiment is that the step S804 specifically includes: Step S901, generating a main node feature matrix, an auxiliary node feature matrix and a new node feature matrix according to the main node, the auxiliary node and the new node.

[0127] In this embodiment, all main nodes in the updated dialogue text feature structure diagram may be extracted, and the number of rows and columns in the matrix may be determined by the position information of each main node in the updated dialogue text feature structure diagram, and the elements in the matrix may be generated by the vector values ​​of each main node, thereby generating a main node feature matrix. All auxiliary nodes in the updated dialogue text feature structure diagram may be extracted, and the number of rows and columns in the matrix may be determined by the position information of each auxiliary node in the updated dialogue text feature structure diagram, and the elements in the matrix may be generated by the vector values ​​of each auxiliary node, thereby generating an auxiliary node feature matrix. All new nodes in the updated dialogue text feature structure diagram may be extracted, and the number of rows and columns in the matrix may be determined by the position information of each new node in the updated dialogue text feature structure diagram, and the elements in the matrix may be generated by the vector values ​​of each new node, thereby generating a new node feature matrix.

[0128] Step S902, concatenating the main node feature matrix, the auxiliary node feature matrix and the new node feature matrix to generate a node feature matrix.

[0129] In this embodiment, the main node feature matrix, the auxiliary node feature matrix and the new node feature matrix may be spliced ​​into one matrix, namely, the node feature matrix.

[0130] Step S903, calculating the reply text vector according to the adjacency variable matrix and the node feature matrix.

[0131] In this embodiment, the dot product calculation may be performed on the adjacency variable matrix and the node feature matrix, and the calculation result is output as the reply text vector.

[0132] Step S904: Generate reply text information and reply voice information according to the reply text vector.

[0133] In this embodiment, the reply text vector can be decoded by the language model to generate a reply text message, and then the reply text message can be converted into a reply voice message by the speech synthesis technology. The reply text message can be visually output through the electronic screen of the AI ​​smart pen holder, and the reply voice message can be output through the speaker of the AI ​​smart pen holder.

[0134] The human-computer interaction method based on the AI ​​smart pen holder provided in the embodiment of the present application quantifies the key information in the conversation by generating a main node feature matrix, an auxiliary node feature matrix and a new node feature matrix, highlights the key attributes of the core content through the main node feature matrix, provides content supplementary details through the auxiliary node feature matrix, and clarifies the latest focus of the current user through the new node feature matrix, thereby improving the efficiency and accuracy of information processing for the conversation content, making the generated reply information more in line with user needs, and improving the fluency and efficiency of human-computer interaction.

[0135] Corresponding to the method of the above embodiment, Fig.10 A structural block diagram of a human-computer interaction device based on an AI smart pen holder provided in an embodiment of the present application is shown. For ease of explanation, only the parts related to the embodiment of the present application are shown. Fig.10 The exemplary human-computer interaction device based on the AI ​​smart pen holder may be the execution subject of the human-computer interaction method based on the AI ​​smart pen holder provided in the aforementioned first embodiment.

[0136] Reference Fig.10 The human-computer interaction device based on the AI ​​smart pen holder includes: The information acquisition module 1010 is used to acquire historical conversation information and voice information of questions input by the user; A text vector generation module 1020, configured to convert the historical conversation information and the question voice information to generate a plurality of historical conversation text vectors and a plurality of question voice text vectors; An initial keyword vector generating module 1030, configured to generate an initial historical conversation keyword vector and an initial question voice keyword vector according to the plurality of historical conversation text vectors, the plurality of question voice text vectors, a preset adjacent region radius, and preset adjacent region vector quantity information; A target keyword vector generating module 1040 is used to generate a target historical conversation keyword vector and a target question voice keyword vector according to the initial historical conversation keyword vector, the initial question voice keyword vector and a plurality of preset feature mapping vectors; A dialogue text feature structure graph construction module 1050 is used to construct a dialogue text feature structure graph according to the historical dialogue text vector and the target historical dialogue keyword vector; and The reply information generation module 1060 is used to generate reply text information and reply voice information according to the dialogue text feature structure diagram and the target question voice text keyword vector.

[0137] The process of each module realizing its own function in the human-computer interaction device based on the AI ​​smart pen holder provided in the embodiment of the present application can be specifically referred to the aforementioned Figure 1 The description of the first embodiment is not repeated here.

[0138] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0139] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0140] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0141] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0142] In addition, in the description of the present specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish descriptions, and cannot be understood as indicating or suggesting relative importance. It should also be understood that although the terms "first", "second", etc. are used to describe various elements in some embodiments of the present application in the text, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, the first table can be named as the second table, and similarly, the second table can be named as the first table without departing from the scope of the various described embodiments. The first table and the second table are both tables, but they are not the same table.

[0143] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0144] The human-computer interaction method based on the AI ​​smart pen holder provided in the embodiment of the present application can be applied to the AI ​​smart pen holder.

[0145] The AI ​​smart pen holder of this embodiment includes: at least one processor and a memory, wherein the memory stores a computer program that can be run on the processor. When the processor executes the computer program, the steps in the above-mentioned human-computer interaction method embodiments based on the AI ​​smart pen holder are implemented, such as Figure 1 Alternatively, when the processor executes the computer program, the functions of each module / unit in the above-mentioned device embodiments are implemented, for example Fig.10 Functions of modules 1010 to 1060 are shown.

[0146] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0147] In some embodiments, the memory may be an internal storage unit of the AI ​​smart pen holder, such as a hard disk or memory of the AI ​​smart pen holder. The AI ​​smart pen holder may also be an external storage device of the AI ​​smart pen holder, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the AI ​​smart pen holder. Furthermore, the memory may include both an internal storage unit and an external storage device of the AI ​​smart pen holder. The memory is used to store operating systems, applications, boot loaders, data, and other programs, such as program codes of computer programs. The memory may also be used to temporarily store data that has been sent or is to be sent.

[0148] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0149] An embodiment of the present application also provides an AI smart pen holder, which includes at least one memory, at least one processor, and a computer program stored in the at least one memory and executable on the at least one processor. When the processor executes the computer program, the AI ​​smart pen holder implements the steps in any of the above-mentioned method embodiments.

[0150] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0151] An embodiment of the present application provides a computer program product. When the computer program product runs on an AI smart pen holder, the AI ​​smart pen holder can implement the steps in the above-mentioned method embodiments when executing.

[0152] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, as well as software distribution media, etc.

[0153] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0154] Those of ordinary skill in the art will appreciate that the units 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 to be beyond the scope of this application.

[0155] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0156] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A human-computer interaction method based on AI smart pen holder, characterized in that: include: Obtain historical conversation information and voice information of questions input by users; Converting the historical conversation information and the question voice information to generate a plurality of historical conversation text vectors and a plurality of question voice text vectors; Generate an initial historical conversation keyword vector and an initial question voice keyword vector according to the plurality of historical conversation text vectors, the plurality of question voice text vectors, a preset adjacent area radius, and preset adjacent area vector quantity information; Generate a target historical conversation keyword vector and a target question voice keyword vector according to the initial historical conversation keyword vector, the initial question voice keyword vector and a plurality of preset feature mapping vectors; Constructing a conversation text feature structure diagram according to the historical conversation text vector and the target historical conversation keyword vector; According to the dialogue text feature structure diagram and the target question voice text keyword vector, a reply text message and a reply voice message are generated.

2. The human-computer interaction method based on the AI ​​smart pen holder according to claim 1, characterized in that: The step of generating an initial historical conversation keyword vector and an initial question voice keyword vector according to the plurality of historical conversation text vectors, the plurality of question voice text vectors, a preset adjacent area radius, and preset adjacent area vector quantity information specifically includes: Generate a human-computer interaction text vector set according to the plurality of historical conversation text vectors and the plurality of question voice text vectors; the human-computer interaction text vector set includes a plurality of human-computer interaction text vectors; Randomly extracting elements of the human-computer interaction text vector set to obtain a human-computer interaction text center vector; Calculating the Euclidean distance between each of the human-computer interaction text vectors and the human-computer interaction text center vector to obtain the interaction text vector distance; When the interactive text vector distance is less than or equal to a preset adjacent area radius, the human-computer interactive text vector corresponding to the interactive text vector distance is used as the human-computer interactive text adjacent area vector; An initial historical dialogue keyword vector and an initial question voice keyword vector are generated according to the human-computer interaction text adjacent region vector and preset adjacent region vector quantity information.

3. The human-computer interaction method based on the AI ​​smart pen holder as claimed in claim 2, characterized in that: The step of generating an initial historical dialogue keyword vector and an initial question voice keyword vector according to the human-computer interaction text adjacent region vector and preset adjacent region vector quantity information specifically includes: Counting the number of vectors in adjacent regions of the human-computer interaction text to obtain information on the number of vectors in the neighborhood of the interaction text; Determine whether the interactive text neighborhood vector quantity information is greater than or equal to the preset adjacent area vector quantity information; If yes, then determining the center vector of the human-computer interaction text as the core vector of the human-computer interaction text; If not, determining the human-computer interaction text center vector as a human-computer interaction text non-core vector; Counting the number of the human-computer interaction text core vectors and the number of the human-computer interaction text non-core vectors to obtain the number of interactive text core vectors and the number of interactive text non-core vectors; Determine whether the sum of the number of interactive text core vectors and the number of interactive text non-core vectors is equal to the number of elements in the human-computer interactive text vector set; If yes, generating an initial historical dialogue keyword vector and an initial question voice keyword vector according to the human-computer interaction text core vector and the human-computer interaction text non-core vector; If not, obtaining a remaining interactive text vector set according to the human-computer interaction text vector set, the human-computer interaction text core vector, and the human-computer interaction text non-core vector; The remaining interactive text vector set is used as a human-computer interaction text vector set, and the step of randomly extracting elements of the human-computer interaction text vector set to obtain a human-computer interaction text center vector is returned.

4. The human-computer interaction method based on the AI ​​smart pen holder as claimed in claim 3, characterized in that: The step of generating an initial historical dialogue keyword vector and an initial question voice keyword vector according to the human-computer interaction text core vector and the human-computer interaction text non-core vector specifically includes: Calculating the Euclidean distance between each non-core vector of the human-computer interaction text and each core vector of the human-computer interaction text to obtain a confirmation distance of a non-core point of the interaction text; Determine whether the confirmation distance of the non-core point of the interactive text is less than or equal to a preset adjacent area radius; If yes, the human-computer interaction text non-core vector corresponding to the interaction text non-core point confirmation distance is used as the human-computer interaction text edge point vector; If not, the non-core vector of the human-computer interaction text corresponding to the non-core point confirmation distance of the interaction text is used as a redundant point vector of the human-computer interaction text; Generate multiple interactive text keyword vectors according to the human-computer interactive text core vector and the human-computer interactive text edge point vector; The interactive text keyword vector is subjected to numerical separation and extraction processing to generate an initial historical conversation keyword vector and an initial question voice keyword vector.

5. The human-computer interaction method based on the AI ​​smart pen holder according to claim 1, characterized in that: The plurality of preset feature mapping vectors include a preset search mapping vector, a preset index mapping vector and a preset content mapping vector; The step of generating a target historical conversation keyword vector and a target question voice keyword vector according to the initial historical conversation keyword vector, the initial question voice keyword vector and a plurality of preset feature mapping vectors specifically includes: Calculate the initial historical conversation keyword retrieval vector, the initial historical conversation keyword index vector, the initial historical conversation keyword content vector, the initial question voice keyword retrieval vector, the initial question voice keyword index vector, and the initial question voice keyword content vector according to the initial historical conversation keyword vector, the initial question voice keyword vector, the preset retrieval mapping vector, the preset index mapping vector, and the preset content mapping vector; Calculating an initial historical conversation keyword query variable according to the initial historical conversation keyword retrieval vector and the initial historical conversation keyword index vector; Calculating an initial question voice keyword query variable according to the initial question voice keyword retrieval vector and the initial question voice keyword index vector; Standardizing the initial historical conversation keyword query variable and the initial question voice keyword query variable to obtain an initial historical keyword query scaling variable and an initial question keyword query scaling variable; Calculating a target historical conversation keyword vector according to the initial historical keyword query scaling variable and the initial historical conversation keyword content vector; A target question voice keyword vector is calculated according to the initial question keyword query scaling variable and the initial question voice keyword content vector.

6. The human-computer interaction method based on the AI ​​smart pen holder according to claim 1, characterized in that: The step of constructing a dialogue text feature structure diagram according to the historical dialogue text vector and the target historical dialogue keyword vector specifically includes: Taking the target historical conversation keyword vector as the main node; Obtaining a historical conversation auxiliary text vector according to the historical conversation text vector and the target historical conversation keyword vector; Using the historical conversation auxiliary text vector as an auxiliary node; Calculating the similarity between each of the primary nodes and the similarity between each of the secondary nodes; Calculate the similarity between each of the primary nodes and each of the secondary nodes to obtain the similarity between the primary and secondary nodes; When the similarity between the master nodes is greater than a preset similarity threshold, edges between the master nodes are generated; When the similarity between the auxiliary nodes is greater than a preset similarity threshold, edges between the auxiliary nodes are generated; When the similarity between the primary and secondary nodes is greater than a preset similarity threshold, an edge between the primary and secondary nodes is generated; A conversation text feature structure graph is constructed based on the main nodes, auxiliary nodes, edges between main nodes, edges between auxiliary nodes, and edges between main and auxiliary nodes.

7. The human-computer interaction method based on the AI ​​smart pen holder according to claim 6, characterized in that: The step of generating reply text information and reply voice information according to the dialogue text feature structure diagram and the target question voice text keyword vector specifically includes: Taking the target question speech text keyword vector as a new node; Randomly inserting the new node into the dialogue text feature structure graph to obtain an updated dialogue text feature structure graph; Calculate the similarity between the new node and the main node to obtain the new main node similarity; Calculating the similarity between the new node and the auxiliary node to obtain the new auxiliary node similarity; When the similarity of the new master node is greater than a preset similarity threshold, an edge between the new node and the master node is generated; When the similarity of the new auxiliary node is greater than a preset similarity threshold, an edge between the new node and the auxiliary node is generated; Reply text information and reply voice information are generated according to the new node, the main node, the auxiliary node, the edge between the main nodes, the edge between the auxiliary nodes, the edge between the main and auxiliary nodes, the edge between the new node and the main node, and the edge between the new node and the auxiliary node.

8. The human-computer interaction method based on the AI ​​smart pen holder according to claim 7, characterized in that: The step of generating a reply text message and a reply voice message according to the new node, the main node, the auxiliary node, the edge between the main node, the edge between the auxiliary nodes, the edge between the main and auxiliary nodes, the edge between the new node and the main node, and the edge between the new node and the auxiliary node specifically includes: Counting the number of new nodes, the number of primary nodes, and the number of secondary nodes; Calculate the total number of nodes according to the number of new nodes, the number of primary nodes, and the number of secondary nodes; Generate an adjacency variable matrix according to the total number of nodes, edges between main nodes, edges between auxiliary nodes, edges between main and auxiliary nodes, edges between new nodes and main nodes, and edges between new nodes and auxiliary nodes; Reply text information and reply voice information are generated according to the adjacency variable matrix, the main node, the auxiliary node and the new node.

9. The human-computer interaction method based on the AI ​​smart pen holder according to claim 8, characterized in that: The step of generating a reply text message and a reply voice message according to the adjacency variable matrix, the main node, the auxiliary node and the new node specifically includes: Generate a main node feature matrix, an auxiliary node feature matrix and a new node feature matrix according to the main node, the auxiliary node and the new node; The main node feature matrix, the auxiliary node feature matrix and the new node feature matrix are concatenated to generate a node feature matrix; Calculate the reply text vector according to the adjacency variable matrix and the node feature matrix; Generate reply text information and reply voice information according to the reply text vector.

10. A human-computer interaction device based on AI smart pen holder, characterized in that: include: The information acquisition module is used to obtain historical conversation information and voice information of questions input by users; A text vector generation module, used to convert the historical conversation information and question voice information to generate a plurality of historical conversation text vectors and a plurality of question voice text vectors; An initial keyword vector generation module, used to generate an initial historical conversation keyword vector and an initial question voice keyword vector according to the plurality of historical conversation text vectors, the plurality of question voice text vectors, a preset adjacent area radius, and preset adjacent area vector quantity information; A target keyword vector generation module, used to generate a target historical conversation keyword vector and a target question voice keyword vector according to the initial historical conversation keyword vector, the initial question voice keyword vector and a plurality of preset feature mapping vectors; A dialogue text feature structure graph construction module, used to construct a dialogue text feature structure graph according to the historical dialogue text vector and the target historical dialogue keyword vector; as well as The reply information generation module is used to generate reply text information and reply voice information according to the dialogue text feature structure diagram and the target question voice text keyword vector.

Images