Intelligent interaction method and system based on AIGC

Abstract

The present application relates to the technical field of intelligent interaction, in particular to an intelligent interaction method and system based on AIGC, the present application splits the input text of the first interaction into interaction vocabulary, then analyzes the term description depth level of the interaction vocabulary, and analyzes the contribution difference between the interaction vocabulary and the effective interaction vocabulary, obtains the interaction depth of the interaction vocabulary, and then sets the initial sampling weight, activates the expert parameters, and obtains the output text of the first interaction; then according to the input text of the second interaction, the change tendency and the change degree of the user's interaction demand are analyzed, the adjustment sampling weight of the second interaction is obtained, the activated expert parameters are inhibited, and the deformation text of the second interaction is output. The present application can effectively improve the matching degree of the output depth knowledge and the actual interaction stage of the user, ensure that the AIGC interaction content is more professional, and optimize the response time of the interaction, so as to maintain the sense of immersion.

Description

Technical Field

[0001] This invention relates to the field of intelligent interaction technology, and more specifically to intelligent interaction methods and systems based on AIGC. Background Technology

[0002] AIGC is an interactive method that outputs AI-generated content, involving multi-dimensional technical implementation and user experience optimization. Its core lies in achieving human-computer interaction through intelligent, personalized, and natural methods. The actual interactive scenario content serves as an activation evaluation of the corresponding expert parameters in the model. By controlling the proportion of activated parameters and the level of expertise, the output results correspond to the user's input interaction requests, thereby optimizing the actual application experience of the interaction.

[0003] Under the changing requirements of domain-specific intelligent agents, accurate evaluation of specific domains and cross-domain integration resulting from user interaction needs require evaluating the user's actual needs based on the user's actual interactive input request text. However, in practice, due to the user's lack of understanding of the search content or the lack of input information, AIGC cannot accurately understand the user's actual needs, resulting in discrepancies between the output interactive text and the user's subjective intent.

[0004] In existing technologies, expert parameters are typically activated by combining the keywords input by the user during the interaction with the input context. However, the leap in understanding that occurs during the initial interaction can lead to a mismatch between the output depth knowledge and the actual stage of the user's interaction, resulting in a waste of expert parameter activation. Summary of the Invention

[0005] To address the above-mentioned technical problems, the present invention aims to provide an intelligent interaction method and system based on AIGC.

[0006] According to a first aspect of the present invention, an AIGC-based intelligent interaction method is provided, the specific technical solution of which is as follows:

[0007] Based on the AIGC platform, the input text of the user's first interaction is obtained, and the input text is broken down into interactive words;

[0008] Analyze the terminology description depth hierarchy of the interactive vocabulary to obtain the depth contribution of each interactive vocabulary;

[0009] Based on the depth contribution, effective interaction words are determined, and the contribution difference between the interaction words and the effective interaction words is analyzed to obtain the interaction depth of the interaction words.

[0010] Based on the interaction depth, initial sampling weights are set, and expert parameters are activated in a targeted manner to obtain the output text of the first interaction;

[0011] If the user interacts again based on the output text, the changes in the user's interaction needs and the degree of change are analyzed according to the input text of the second interaction. The adjusted sampling weights for the second interaction are obtained, the activated expert parameters are suppressed in a targeted manner, and the deformed text of the second interaction is output.

[0012] In some embodiments of the present invention, the terminological description depth hierarchy of the interactive vocabulary is analyzed to obtain the depth contribution of each interactive vocabulary, including:

[0013] In the domain knowledge graph established within the AIGC platform, the corresponding interactive words that are the same as the interactive words are located, and the distance from the corresponding interactive words to the domain name master node of the domain knowledge graph is extracted to obtain the term description depth level corresponding to the interactive words.

[0014] For each of the interactive words, the maximum term description depth level corresponding to it in the domain knowledge graph is extracted, and combined with the term description depth level corresponding to the interactive word, the depth contribution of each interactive word is obtained.

[0015] In some embodiments of the present invention, determining effective interaction vocabulary based on the depth contribution includes:

[0016] Obtain the median of the depth contribution of all the aforementioned interaction terms;

[0017] The interaction words corresponding to the depth contributions that are higher than the median are identified as valid interaction words.

[0018] In some embodiments of the present invention, analyzing the contribution difference between the interactive vocabulary and the effective interactive vocabulary to obtain the interaction depth of the interactive vocabulary includes:

[0019] Calculate the difference in depth contribution between the effective interaction words and the interaction words to obtain the contribution difference between the interaction words and the effective interaction words;

[0020] Based on the order of the input text, sort all the interactive words to obtain an interactive word sequence;

[0021] In the sequence of interactive words, the distance between the interactive words and the effective interactive words is calculated to obtain a distance parameter;

[0022] The interaction depth of the interactive vocabulary is obtained by combining the contribution difference and the distance parameter.

[0023] In some embodiments of the present invention, the interaction depth of the interactive vocabulary is obtained by combining the contribution difference and the distance parameter, including:

[0024] By combining the contribution differences and the distance parameter, partial depth contributions of all the effective interaction words are obtained;

[0025] By traversing all the valid interactive words in the input text, calculating the sum of the partial depth contributions, and combining this with the total number of interactive words, the interaction depth of the interactive words is obtained.

[0026] In some embodiments of the present invention, based on the interaction depth, initial sampling weights are set, and expert parameters are activated in a targeted manner to obtain the output text of the first interaction, including:

[0027] In the domain knowledge graph established within the AIGC platform, obtain the domain name master nodes corresponding to all the interactive words, and activate the expert sub-networks corresponding to the domain name master nodes;

[0028] The interaction depth is normalized using softsign to obtain the normalized interaction depth.

[0029] The initial sampling weight is set to a percentage of the normalized interaction depth. The output string of all activated expert subnetworks is then taken as the result of the first percentage of the normalized interaction depth to obtain the output text of the first interaction.

[0030] In some embodiments of the present invention, analyzing the changing trends of user interaction needs based on the input text of the subsequent interaction includes:

[0031] Based on the input text of the second interaction, extract the second effective interaction words of the second interaction, and obtain the domain name master node corresponding to the second effective interaction words;

[0032] The total number of times the domain name master node corresponding to each of the aforementioned effective interaction words appears in all interaction processes is counted. Combined with the total number of interactions, the domain emphasis evaluation of the domain name master node corresponding to each of the aforementioned effective interaction words is obtained.

[0033] The number of times the domain name master node corresponding to each of the re-valid interaction words appears in the re-interaction is counted. Combined with the total number of all interaction words in the re-interaction, the query quantity parameter of the domain name master node corresponding to each of the re-valid interaction words is obtained.

[0034] By combining the domain-specific evaluation and the number of inquiries, the changing trends of users' interaction needs can be obtained.

[0035] In some embodiments of the present invention, analyzing the degree of change in the user's interaction needs based on the input text of the subsequent interaction includes:

[0036] Obtain a partial depth contribution of the re-valid interaction vocabulary, and obtain the domain name master node corresponding to the re-valid interaction vocabulary;

[0037] Based on the partial depth contribution of the re-effective interaction vocabulary and the domain name master node, the interaction vector of the re-interaction is obtained;

[0038] Based on the input text of the previous interaction, obtain the interaction vector of the previous interaction;

[0039] By analyzing the similarity between the interaction vectors corresponding to the second interaction and the previous interaction, the degree of change in the user's interaction needs between the second interaction and the previous interaction can be obtained.

[0040] According to a second aspect of the present invention, an AIGC-based intelligent interactive system is provided, comprising: a memory, a processor, and a display, wherein:

[0041] The memory is used to store program code;

[0042] The processor is configured to read program code stored in the memory and execute the method described in the first aspect of the present invention.

[0043] The display is used to show the user's input text and the output text after processing the user's input text.

[0044] In some embodiments of the present invention, the processor includes:

[0045] The interactive vocabulary acquisition module is used to acquire the input text of the user's first interaction based on the AIGC platform, and to break down the input text into interactive vocabulary.

[0046] An interaction depth analysis module is used to analyze the terminology description depth level of the interaction words to obtain the depth contribution of each interaction word; and to determine the effective interaction words based on the depth contribution, and analyze the contribution difference between the interaction words and the effective interaction words to obtain the interaction depth of the interaction words.

[0047] The output text acquisition module is used to set initial sampling weights based on the interaction depth, perform targeted activation of expert parameters, and obtain the output text of the first interaction.

[0048] The output text optimization module is used to analyze the changing trend and degree of the user's interaction needs based on the input text of the second interaction if the user interacts again based on the output text, obtain the adjustment sampling weight of the second interaction, suppress the activated expert parameters in a targeted manner, and output the deformed text of the second interaction.

[0049] Compared with existing technologies, the AIGC-based intelligent interaction method and system provided by this invention have the following beneficial effects:

[0050] This invention first breaks down the input text into interactive words that can be analyzed by the model, providing foundational data for subsequent analysis. Then, by analyzing the terminological depth of the interactive words, the depth contribution of each interactive word is obtained. Further analysis of the contribution differences between interactive words and effective interactive words yields the interaction depth of the interactive words. In other words, by analyzing the semantic contribution of interactive words, the input result of the current interactive words can be accurately reflected, thereby enabling the evaluation of the output process. Then, based on the interaction depth of the sentences, the sampling ratio of activation parameters is determined to obtain the output text. Next, by analyzing the changes in the interactive needs reflected in the re-input text compared to the previous input text, the output content is re-evaluated after deepening the user interaction level, achieving targeted suppression of activated expert parameters and outputting modified text for further interaction. The technical solution provided by this invention effectively improves the matching degree between the output depth knowledge and the actual user interaction stage, ensuring that AIGC interactive content is more professional while optimizing the interactive experience. Attached Figure Description

[0051] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a basic flowchart illustrating an AIGC-based intelligent interaction method provided in one embodiment of the present invention.

[0053] Figure 2 This is a schematic diagram of the basic components of an AIGC-based intelligent interactive system provided in one embodiment of the present invention. Detailed Implementation

[0054] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the AIGC-based intelligent interaction method and system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Terms such as “comprising,” “including,” or any other variations thereof are intended to cover a non-exclusive inclusion, such that a circuit structure, article, or device comprising a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such article or device. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of additional identical elements in the article or device that includes the element.

[0056] The specific solution of the AIGC-based intelligent interaction method provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0057] Please see Figure 1 This illustrates the basic flow of an AIGC-based intelligent interaction method provided by an embodiment of the present invention.

[0058] like Figure 1 As shown, an embodiment of the present invention provides an intelligent interaction method based on AIGC, which specifically includes:

[0059] S100: Based on the AIGC platform, it acquires the input text of the user's first interaction and breaks down the input text into interactive words.

[0060] The AIGC platform provides a visual display through a graphical user interface (GUI) and includes interactive modules for users. The platform extracts the input text from the user's initial interaction using the GUI. Preprocessing of the input text includes text cleaning and word segmentation. Specifically, a stop word replacement dictionary is used to replace abnormal words in the initial user interaction input text, ensuring that all words in the input text fall within the technical scope of the model training; special characters (excluding special symbols) are removed; and the input text is segmented into interactive words using the precise mode of the jieba word segmentation algorithm.

[0061] This completes the acquisition and preprocessing of the input text from the user's first interaction, yielding interactive vocabulary that can be used for model analysis.

[0062] S200: Analyze the terminological description depth of interactive vocabulary to obtain the depth contribution of each interactive vocabulary.

[0063] A Generative Pre-trained Transformer (GPT) is an advanced artificial intelligence language model that uses deep learning techniques, particularly the Transformer architecture, to understand and generate natural language text. Through pre-training on large amounts of text data, GPT learns language patterns and structures, enabling it to predict and generate coherent and meaningful text content. GPT models can be widely applied to various natural language processing tasks such as text generation, dialogue systems, and automatic summarization. In embodiments of this invention, a pre-trained generative pre-trained transformer model is used as the foundation for the input text processing model.

[0064] When users interact, they usually ask about a specific professional knowledge point to obtain a description of the key technology. Therefore, in the process of interacting with a professional field, users evaluate the output process by assessing the level of terminology input for the professional knowledge point. If the user is very familiar with a certain technical field, the terminology input will be more professional, that is, the input terminology belongs to the deep vocabulary of one or more technical fields.

[0065] Based on the above analysis, in some embodiments of the present invention, the depth contribution of each interactive word is obtained by analyzing the terminological description depth hierarchy of the interactive words. Further, this includes:

[0066] First, the corresponding interaction terms that match the interaction terms are located in the domain knowledge graph established within the AIGC platform. The distance from the corresponding interaction term to the domain name master node in the domain knowledge graph is extracted to obtain the terminology description depth level corresponding to the interaction term. Specifically, the node name and relation name fields of the domain knowledge graph established within the AIGC platform are used to filter the extracted interaction terms, and the corresponding interaction terms are located in the domain knowledge graph. The same corresponding interactive vocabulary; the domain knowledge graph has a tree structure, and the further away from the main node of the domain name, the deeper it is. Therefore, extracting interactive vocabulary... The corresponding interactive words are mapped to the domain name master nodes of the domain knowledge graph. distance That is, the terminology description depth level corresponding to the interactive vocabulary is obtained as follows: For example: Vehicle (domain name master node) - Driving mechanism - Steering unit - Steering wheel (current interaction term) If the distance between the steering wheel and the vehicle is 3, then the depth level of the current interactive word (steering wheel) is 3.

[0067] This analysis focuses on the interactive description methods for interactive terms with higher levels of descriptive depth. Higher-level descriptive terms represent deeper and more specialized technical concepts, resulting in a narrower scope of user interaction. However, the descriptive methods matching higher-level descriptive terms are more specialized than those for lower-level terms, meaning the related terminology levels are close. This indicates that the interaction needs for higher-level descriptive terms require the display of other interactive terms at similar levels. For example, in the phrase "CKF and UKF both calculate the first and second moments of the filter distribution through a set of weighted sampling points and a nonlinear system transformation," the terms "nonlinear system," "filter distribution," and "second moment" have a level of 7. Therefore, if the user's initial input text has many features with similar and higher levels of descriptive depth, the initial interaction depth rating is higher, leading to more accurate activation of expert parameters.

[0068] Therefore, in some embodiments of the present invention, for each interactive word, the maximum term description depth level corresponding to its domain knowledge graph is extracted, and combined with the term description depth level corresponding to the interactive word, the depth contribution of each interactive word is obtained. The specific implementation method is as follows:

[0069] Targeting interactive vocabulary Extract the domain knowledge graph to which it belongs. The corresponding maximum term description depth level (Each domain knowledge graph has its corresponding deepest level); then calculate the interaction vocabulary. The corresponding terminology describes the depth level. Describing the depth level with the maximum term The ratio is:

[0070]

[0071] In the formula, Interactive vocabulary The ratio of the corresponding term description depth level to the maximum term description depth level; Interactive vocabulary The corresponding terminology describes the depth level; Interactive vocabulary Domain-specific knowledge graph The corresponding maximum terminology description depth level.

[0072] And calculate the maximum term description depth level. All interactive words corresponding to the initial input text are mapped to the domain knowledge graph. The mean of the ratios of the distances to the domain name master nodes is:

[0073]

[0074] In the formula, This represents the average ratio of the maximum term description depth level to the term description depth level corresponding to the first interactive input text; Interactive vocabulary Domain-specific knowledge graph The corresponding maximum terminology description depth level; This indicates the first time the text was entered during the interaction. All interactive words To Domain Knowledge Graph Domain name master node Distance (if the domain knowledge graph) If no corresponding interactive word can be located in the code, then it does not exist. ,Right now (Value is 0) This represents the total number of all interactive words corresponding to the initial input text.

[0075] Represents the current domain knowledge graph The term "internal average" describes the user's interaction requests as manifested in the depth hierarchy of a domain knowledge graph. The user's domain preference, i.e., whether the user asks questions that are relevant to the current field.

[0076] Finally, calculate the interaction vocabulary. The corresponding ratio and domain knowledge graph The ratio of the mean of the corresponding ratios yields the interactive vocabulary. The profound contribution is as follows:

[0077]

[0078] In the formula, Interactive vocabulary The depth of contribution is as follows, and the meanings of the other symbols are the same as those in the two equations above.

[0079] Iterate through the interactive words corresponding to the initial interactive input text to obtain the depth contribution of each interactive word.

[0080] S300: Based on the depth contribution, identify the effective interactive words and analyze the contribution difference between the interactive words and the effective interactive words to obtain the interaction depth of the interactive words.

[0081] Based on the order of the input text, all interactive words are sorted to form an interactive word sequence. The depth contribution of interactive words is placed according to the order of the interactive word sequence. The descriptive help of each word at the peak position of each depth contribution is expressed as follows: the greater the distance difference between two adjacent interactive words at high depth contribution positions, the higher the contribution of other interactive words close to it, and the stronger the descriptive ability of the current interactive word for the interactive words with high depth contributions. Furthermore, the number of interactive words contained in the current interactive word sequence is counted to achieve an accurate assessment of the interaction depth of the input text.

[0082] Based on the above analysis, in some embodiments of the present invention, effective interactive words are determined according to their depth contribution, and the contribution difference between interactive words and effective interactive words is analyzed to obtain the interaction depth of the interactive words. Wherein:

[0083] Based on depth contribution, effective interaction words are determined, including: obtaining the median of the depth contribution of all interaction words; and identifying the interaction words with depth contributions higher than the median as effective interaction words.

[0084] The interaction depth of interactive words is obtained by analyzing the contribution difference between interactive words and effective interactive words. This includes: calculating the difference in depth contribution between effective interactive words and interactive words to obtain the contribution difference between interactive words and effective interactive words; sorting all interactive words according to the order of the input text to obtain an interactive word sequence; calculating the distance between interactive words and effective interactive words in the interactive word sequence to obtain a distance parameter; and combining the contribution difference, distance parameter, and other parameters to obtain the interaction depth of interactive words.

[0085] This involves combining contribution differences and distance parameters to obtain the interaction depth of interactive words. Further steps include: first, combining contribution differences and distance parameters to obtain the partial depth contribution of each effective interactive word, thus constructing an effective interactive vocabulary. The formula for calculating part of the depth contribution is:

[0086]

[0087] In the formula, Indicates valid interaction vocabulary Partial depth contribution; Indicates valid interaction vocabulary _profound_contribution; Indicates and effective interaction vocabulary The nearest non-effective interaction word The depth contribution, if the nearest ineffective interaction word. If there is more than one, then the depth of contribution will be considered. and closest (i.e.) The least effective interaction word (the one with the smallest value) ; This indicates the number of valid interactive words in the interactive word sequence. Ineffective interaction words The distance between them.

[0088] Then, iterate through all valid interactive words in the input text, calculate the sum of partial depth contributions, and combine this with the total number of interactive words to obtain the interaction depth of the interactive words:

[0089]

[0090] In the formula, This indicates the interaction depth of the interaction words corresponding to the first interaction; Indicates valid interaction vocabulary Partial depth contribution; This indicates the valid interaction words corresponding to the initial input text. The total quantity; This represents the total number of all interactive words corresponding to the initial input text.

[0091] The larger the value, the longer the text input during the user's first interaction. Weighting contributions to reflect the dilution of deep contributions in long sentences and evaluating the interaction depth of the current input word sequence can more accurately reflect the input results of the current interaction word sequence.

[0092] This concludes the analysis of the semantic contribution of the interactive words corresponding to the initial interactive input text.

[0093] S400: Based on the interaction depth, set the initial sampling weights, perform targeted activation of expert parameters, and obtain the output text of the first interaction.

[0094] After obtaining the interaction depth evaluation results of the interaction words corresponding to the first interaction, targeted activation of expert parameters is performed. Expert parameters refer to the parameters of each sub-network contained in the overall model, that is, the results of training with knowledge parameters from a specific domain. Therefore, the goal is to achieve targeted activation of sub-networks by input parameters, thereby determining the proportion of different professional domain knowledge in the output content. Therefore, the sampling ratio of activation parameters is determined according to the interaction depth of the sentence, thereby further deepening the user interaction level and re-evaluating the output content.

[0095] Based on the above analysis, in some embodiments of the present invention, initial sampling weights are set according to the interaction depth, and expert parameters are activated in a targeted manner to obtain the output text of the first interaction. Specifically, this includes:

[0096] In the domain knowledge graph established within the AIGC platform, obtain the domain name master nodes corresponding to all interactive terms, and activate the expert subnetworks corresponding to the domain name master nodes (the domain name master nodes and expert subnetworks have the same name).

[0097] Deepen the interaction Perform softsign normalization to obtain the normalized interaction depth. ;

[0098] The initial sampling weight is set to the normalized interaction depth. The output string of all activated expert subnetworks is taken as the result of the initial sampling weight (normalized interaction depth) (the total weight is set to the sum of the initial sampling weights of all expert subnetworks to avoid weight overflow caused by setting the fixed weight to 1), and the output text of the first interaction is obtained.

[0099] S500: If the user interacts again based on the output text, the system analyzes the changing trend and degree of the user's interaction needs based on the input text of the second interaction, obtains the adjusted sampling weight for the second interaction, performs targeted suppression on the activated expert parameters, and outputs the deformed text of the second interaction.

[0100] After receiving the output text, users will re-establish a focus on a particular area during subsequent interactive queries. Therefore, the changes in the output results should be evaluated based on the changes in the depth of user interaction in a particular area.

[0101] Therefore, if the user interacts again based on the output text, the algorithm analyzes the changing tendencies and degrees of change in the user's interaction needs based on the input text of the second interaction, obtains the adjusted sampling weights for the second interaction, and specifically suppresses the activated expert parameters to output the modified text for the second interaction. Wherein:

[0102] Based on the input text from subsequent interactions, analyze the changing trends in user interaction needs, including:

[0103] First, based on the input text of the second interaction, extract the effective interaction words for the second interaction and obtain the domain name master node corresponding to the effective interaction words. Specifically, the method for extracting the effective interaction words for the second interaction based on the input text of the second interaction is the same as steps S100 to S300, that is, based on the AIGC platform, obtain the user's second input text for the second interaction and break it down into second interaction words; analyze the terminology description depth level of the second interaction words to obtain the depth contribution of each second interaction word; and determine the effective interaction words corresponding to the second input text based on the depth contribution. Detailed steps are shown in steps S100 to S300 and will not be repeated here. Then, if in the domain knowledge graph... If the corresponding repeated interaction words that are the same as the words used in the repeated effective interaction can be located, then the knowledge graph of that domain can be established. The domain name master node in the code is the domain name master node corresponding to the valid interaction term.

[0104] Next, the effective interaction vocabulary is compared with historical interactions. This involves evaluating the questioning methods. The quantity and variety of domain-specific interactions after each interaction represent the user's more realistic interaction needs. For example, as the level of professional expertise in the interaction deepens, the remaining effective interaction vocabulary shows a gradual decrease in domain name main nodes, with concentrated descriptive words appearing in a very small number of domain name main nodes, indicating that the interaction domain is gradually narrowing and deepening. Therefore, the repetition of domain name main nodes needs to be evaluated first.

[0105] Based on the above analysis, in some embodiments of the present invention, by counting the total number of times the domain name master node corresponding to each re-valid interaction term appears in all interaction processes, and combining this with the total number of interactions, the domain emphasis evaluation of the domain name master node corresponding to each re-valid interaction term is obtained as follows:

[0106]

[0107] In the formula, Indicates further interaction (the first) The domain name master node corresponding to the next valid interaction term (secondary interaction). The field focuses on evaluation; This indicates the domain name master node corresponding to the valid interaction term. In all Total number of occurrences during this interaction process; This represents the total number of interactions the user has made up to the present.

[0108] Furthermore, the number of times the domain name master node corresponding to each subsequent valid interaction term appears in subsequent interactions is counted. Combined with the total number of all interaction terms in subsequent interactions, the query count parameter for the domain name master node corresponding to each subsequent valid interaction term is obtained as follows:

[0109]

[0110] In the formula, Indicates further interaction (the first) The domain name master node corresponding to the next valid interaction term (secondary interaction). The number of queries parameter; This indicates the domain name master node corresponding to the valid interaction term. In the next interaction (currently the first) The number of times it appears in each interaction; This indicates the total number of all words that can be interacted with again in response to the input text.

[0111] In the case of in-depth inquiry, the level of inquiry is deeper, so the inquiry method includes more detailed content in a single field, indicating that the user's inquiry is more professional, and therefore more in-depth terminology in the same field is needed for description.

[0112] Finally, combining the domain-specific evaluation and inquiry quantity parameters, the changing trend of user interaction needs is as follows:

[0113]

[0114] In the formula, Indicates further interaction (the first) The user's interaction needs under the second interaction (for the domain name master node) The tendency of change; Indicates further interaction (the first) The domain name master node corresponding to the next valid interaction term (secondary interaction). The field focuses on evaluation; Indicates further interaction (the first) The domain name master node corresponding to the next valid interaction term (secondary interaction). The number of queries parameter.

[0115] Based on the input text from subsequent interactions, analyze the degree of change in the user's interaction needs, including:

[0116] First, obtain the partial depth contribution of the re-effective interaction terms and the domain name master nodes corresponding to the re-effective interaction terms. The specific method for obtaining the partial depth contribution of the re-effective interaction terms is the same as in step S300, that is, combining the contribution difference and distance parameters to obtain the partial depth contribution of all re-effective interaction terms. Detailed steps are described in step S300 and will not be repeated here. Then, if in the domain knowledge graph... If the corresponding repeated interaction words that are the same as the words used in the repeated effective interaction can be located, then the knowledge graph of that domain can be established. The domain name master node in the context is the domain name master node corresponding to the valid interaction term. .

[0117] Then, based on the partial depth contribution of the effective interaction vocabulary and the domain name master node, the comparison point is set as... ,in Indicates the domain name master node, Indicates a word that represents another valid interaction. The partial depth contribution; the interaction vector of the second interaction is obtained by arranging the comparison points in the order of the second interaction words of the re-input text, denoted as .

[0118] Similarly, based on the input text of the previous interaction (the previous interaction related to the next interaction level), the interaction vector of the previous interaction is obtained, denoted as... ;

[0119] Finally, by analyzing the similarity between the interaction vectors corresponding to the second interaction and the previous interaction, we can obtain the degree of change in the user's interaction needs between the second interaction and the previous interaction:

[0120]

[0121] In the formula, This indicates that the user is interacting again (the first time). The first interaction) and the previous interaction (the second interaction) The degree of change in interaction needs (multiple interactions); Represents the interaction vector of the previous interaction related to the current interaction level. Interaction vectors with subsequent interactions Cosine similarity between them.

[0122] The current query method is determined by the query order among interaction vectors. Specifically, the similarity between interaction vectors is used as the background activation parameter for the current query, taking into account the query result deviation (similarity between interaction vectors). This is further combined with the domain name master node. The tendency of change, that is, combining the domain name master node The domain focuses on evaluating and querying the number of parameters to obtain the domain name master node. The sampling weights are adjusted in the subsequent interaction as follows:

[0123]

[0124] In the formula, Indicates the domain name master node In the next interaction (the first) Adjusting sampling weights under (secondary interaction); Indicates further interaction (the first) The user's interaction needs under the second interaction (for the domain name master node) The tendency of change; This indicates that the user is interacting again (the first time). The first interaction) and the previous interaction (the second interaction) The degree of change in interaction needs (multiple interactions).

[0125] By analyzing the two proportional values ​​( and The sum of 1 and -1 is used to complete the mapping from logarithmic values ​​to 1, thereby ensuring that the activation level of expert parameters is similar when the questioning method does not change much, and stabilizing the output of the response results in the interactive process.

[0126] Similarly, the main nodes of each domain name are obtained in the subsequent interaction (the first... Adjust the sampling weights in the next interaction. Then, based on the adjusted sampling weights, set further adjustment sampling weights to specifically suppress the activated expert parameters, and output the transformed text for the next interaction. The specific operation method is the same as step S400, that is, adjust the sampling weights... Perform softsign normalization to obtain the normalized interaction depth. Set the initial sampling weight to 100% normalized interaction depth. For all activated expert subnetworks, take the first percent of the normalized interaction depth from the output strings. The result (the total weight is set to the sum of the initial sample weights of all expert subnetworks to avoid weight overflow caused by fixing the weight to 1) outputs the transformed text for further interaction.

[0127] Finally, the interactive results are visualized. Specifically:

[0128] After the interactive statements are generated, they are streamed in real time: Server-Sent Events (SSE) are used to support pushing text word by word / sentence by sentence, sending the text to the client's memory.

[0129] The message history area is displayed as follows:

[0130] a) User input and model response are displayed side-by-side (similar to a chat interface);

[0131] b) User messages (right side): light background + right alignment;

[0132] c) Model response (left side): Dark background + left alignment + fixed maximum width.

[0133] The input control area layout consists of: a text input box (supporting multiple lines), a send button, and additional function buttons (such as clear history and settings).

[0134] A status indicator is used to indicate whether the output has been completed.

[0135] Loading animations (such as progress bars, rotating icons), and error message areas.

[0136] Based on the same inventive concept as the above method, this embodiment also provides an AIGC-based intelligent interaction system.

[0137] Please see Figure 2 This illustrates the basic components of an AIGC-based intelligent interactive system provided by an embodiment of the present invention.

[0138] like Figure 2 As shown, an AIGC-based intelligent interactive system includes a memory 10, a processor 20, and a display 30, wherein:

[0139] Memory 10 is used to store program code.

[0140] The processor 20 reads the program code stored in the memory 10 and executes it based on the AIGC platform. It acquires the input text of the user's first interaction and breaks it down into interactive words. It analyzes the terminology description depth of the interactive words to obtain the depth contribution of each word. Based on the depth contribution, it determines the effective interactive words and analyzes the contribution difference between the interactive words and the effective interactive words to obtain the interaction depth of the interactive words. Based on the interaction depth, it sets initial sampling weights, performs targeted activation of expert parameters, and obtains the output text of the first interaction. If the user interacts again based on the output text, it analyzes the changing tendency and degree of the user's interaction needs based on the input text of the second interaction to obtain adjusted sampling weights for the second interaction, performs targeted suppression of the activated expert parameters, and outputs the transformed text of the second interaction.

[0141] Display 30 is used to display the user's input text and the output text after processing the user's input text.

[0142] Furthermore, the processor 20 includes an interactive vocabulary acquisition module 21, an interactive depth analysis module 22, an output text acquisition module 23, and an output text optimization module 24. Wherein:

[0143] The interactive vocabulary acquisition module 21 is used to acquire the input text of the user's first interaction based on the AIGC platform, and to break down the input text into interactive vocabulary;

[0144] The interaction depth analysis module 22 is used to analyze the terminology description depth level of interactive words to obtain the depth contribution of each interactive word; and to determine the effective interactive words based on the depth contribution, and to analyze the contribution difference between interactive words and effective interactive words to obtain the interaction depth of interactive words.

[0145] The output text acquisition module 23 is used to set the initial sampling weight according to the interaction depth, perform targeted activation of expert parameters, and obtain the output text of the first interaction.

[0146] The output text optimization module 24 is used to analyze the changing trend and degree of the user's interaction needs based on the input text of the second interaction if the user interacts again based on the output text, obtain the adjustment sampling weight of the second interaction, suppress the activated expert parameters in a targeted manner, and output the deformed text of the second interaction.

[0147] The display 30 includes a message history display area, an input control area, a status indicator, and an animation loading area. Among these:

[0148] The layout of the message history display area is as follows:

[0149] a) User input and model response are displayed side-by-side (similar to a chat interface);

[0150] b) User messages (right side): light background + right alignment;

[0151] c) Model response (left side): Dark background + left alignment + fixed maximum width.

[0152] The input control area layout consists of: a text input box (supporting multiple lines), a send button, and additional function buttons (such as clear history and settings).

[0153] A status indicator is used to indicate whether the output has been completed.

[0154] The animation loading area includes a progress bar, a rotating icon, and an error message sub-area.

[0155] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0156] The various embodiments in this specification are described in a progressive manner. For the same or similar parts between the various embodiments, please refer to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. An intelligent interaction method based on AIGC, characterized in that, The method includes: Based on the AIGC platform, the input text of the user's first interaction is obtained, and the input text is broken down into interactive words; Analyze the terminology description depth hierarchy of the interactive vocabulary to obtain the depth contribution of each interactive vocabulary; Based on the depth contribution, effective interaction words are determined, and the contribution difference between the interaction words and the effective interaction words is analyzed to obtain the interaction depth of the interaction words. Based on the interaction depth, initial sampling weights are set, and expert parameters are activated in a targeted manner to obtain the output text of the first interaction; If the user interacts again based on the output text, the user's interaction needs are analyzed and their degree of change is determined based on the input text of the second interaction. The adjusted sampling weights for the second interaction are obtained, the activated expert parameters are suppressed in a targeted manner, and the deformed text of the second interaction is output. The method for obtaining the depth contribution of each of the interactive words includes: In the domain knowledge graph established within the AIGC platform, the corresponding interactive words that are the same as the interactive words are located, and the distance from the corresponding interactive words to the domain name master node of the domain knowledge graph is extracted to obtain the term description depth level corresponding to the interactive words. For each interactive term, extract the maximum term description depth level corresponding to it in the domain knowledge graph, and combine it with the term description depth level corresponding to the interactive term; calculate the interactive term. Corresponding terminology description depth level Describing the depth level with the maximum term ratio ,for: ; Calculate the maximum term description depth level All interactive words corresponding to the initial input text are mapped to the domain knowledge graph. The mean of the ratios of the distances to the domain name master node ,for: ; In the formula, Interactive vocabulary Domain-specific knowledge graph The corresponding maximum terminology description depth level; This indicates the first time the text was entered during the interaction. All interactive words To Domain Knowledge Graph Domain name master node The distance; This represents the total number of all interactive words corresponding to the initial interactive input text. Computational interactive vocabulary The corresponding ratio and domain knowledge graph The ratio of the mean of the corresponding ratios Get interactive vocabulary Its profound contributions.

2. The AIGC-based intelligent interaction method according to claim 1, characterized in that, Based on the depth contribution, valid interaction vocabulary is determined, including: Obtain the median of the depth contribution of all the aforementioned interaction terms; The interaction words corresponding to the depth contributions that are higher than the median are identified as valid interaction words.

3. The intelligent interaction method based on AIGC according to claim 2, characterized in that, Analyzing the contribution difference between the interactive vocabulary and the effective interactive vocabulary yields the interaction depth of the interactive vocabulary, including: Calculate the difference in depth contribution between the effective interaction words and the interaction words to obtain the contribution difference between the interaction words and the effective interaction words; Based on the order of the input text, sort all the interactive words to obtain an interactive word sequence; In the sequence of interactive words, the distance between the interactive words and the effective interactive words is calculated to obtain a distance parameter; The interaction depth of the interactive vocabulary is obtained by combining the contribution difference and the distance parameter.

4. The AIGC-based intelligent interaction method according to claim 3, characterized in that, Combining the contribution difference and the distance parameter, the interaction depth of the interactive vocabulary is obtained, including: By combining the contribution differences and the distance parameter, partial depth contributions of all the effective interaction words are obtained; By traversing all the valid interactive words in the input text, calculating the sum of the partial depth contributions, and combining this with the total number of interactive words, the interaction depth of the interactive words is obtained.

5. The intelligent interaction method based on AIGC according to claim 1, characterized in that, Based on the interaction depth, initial sampling weights are set, and expert parameters are activated in a targeted manner to obtain the output text of the first interaction, including: In the domain knowledge graph established within the AIGC platform, obtain the domain name master nodes corresponding to all the interactive words, and activate the expert sub-networks corresponding to the domain name master nodes; The interaction depth is normalized using softsign to obtain the normalized interaction depth. The initial sampling weight is set to a percentage of the normalized interaction depth. The output string of all activated expert subnetworks is then taken as the result of the first percentage of the normalized interaction depth to obtain the output text of the first interaction.

6. The AIGC-based intelligent interaction method according to claim 3, characterized in that, Based on the input text from the subsequent interaction, analyze the changing trends in the user's interaction needs, including: Based on the input text of the second interaction, extract the second effective interaction words of the second interaction, and obtain the domain name master node corresponding to the second effective interaction words; The total number of times the domain name master node corresponding to each of the aforementioned effective interaction words appears in all interaction processes is counted. Combined with the total number of interactions, the domain emphasis evaluation of the domain name master node corresponding to each of the aforementioned effective interaction words is obtained. The number of times the domain name master node corresponding to each of the re-valid interaction words appears in the re-interaction is counted. Combined with the total number of all interaction words in the re-interaction, the query quantity parameter of the domain name master node corresponding to each of the re-valid interaction words is obtained. By combining the domain-specific evaluation and the number of inquiries, the changing trends of users' interaction needs can be obtained.

7. The intelligent interaction method based on AIGC according to claim 6, characterized in that, Based on the input text from the subsequent interaction, analyze the degree of change in the user's interaction needs, including: Obtain a partial depth contribution of the re-valid interaction vocabulary, and obtain the domain name master node corresponding to the re-valid interaction vocabulary; Based on the partial depth contribution of the re-effective interaction vocabulary and the domain name master node, the interaction vector of the re-interaction is obtained; Based on the input text of the previous interaction, obtain the interaction vector of the previous interaction; By analyzing the similarity between the interaction vectors corresponding to the second interaction and the previous interaction, the degree of change in the user's interaction needs between the second interaction and the previous interaction can be obtained.

8. An intelligent interactive system based on AIGC, characterized in that, The system includes: a memory, a processor, and a display, wherein: The memory is used to store program code; The processor is configured to read program code stored in the memory and execute the method as described in any one of claims 1 to 7; The display is used to show the user's input text and the output text after processing the user's input text.

9. The intelligent interactive system based on AIGC according to claim 8, characterized in that, The processor includes: The interactive vocabulary acquisition module is used to acquire the input text of the user's first interaction based on the AIGC platform, and to break down the input text into interactive vocabulary. An interaction depth analysis module is used to analyze the terminology description depth level of the interaction words to obtain the depth contribution of each interaction word; and to determine the effective interaction words based on the depth contribution, and analyze the contribution difference between the interaction words and the effective interaction words to obtain the interaction depth of the interaction words. The output text acquisition module is used to set initial sampling weights based on the interaction depth, perform targeted activation of expert parameters, and obtain the output text of the first interaction. The output text optimization module is used to analyze the changing trend and degree of the user's interaction needs based on the input text of the second interaction if the user interacts again based on the output text, obtain the adjustment sampling weight of the second interaction, suppress the activated expert parameters in a targeted manner, and output the deformed text of the second interaction. The method for obtaining the depth contribution of each of the interactive words includes: In the domain knowledge graph established within the AIGC platform, the corresponding interactive words that are the same as the interactive words are located, and the distance from the corresponding interactive words to the domain name master node of the domain knowledge graph is extracted to obtain the term description depth level corresponding to the interactive words. For each interactive term, extract the maximum term description depth level corresponding to it in the domain knowledge graph, and combine it with the term description depth level corresponding to the interactive term; calculate the interactive term. Corresponding terminology description depth level Describing the depth level with the maximum term ratio ,for: ; Calculate the maximum term description depth level All interactive words corresponding to the initial input text are mapped to the domain knowledge graph. The mean of the ratios of the distances to the domain name master node ,for: ; In the formula, Interactive vocabulary Domain-specific knowledge graph The corresponding maximum terminology description depth level; This indicates the first time the text was entered during the interaction. All interactive words To Domain Knowledge Graph Domain name master node The distance; This represents the total number of all interactive words corresponding to the initial input text; it also calculates the number of interactive words. The corresponding ratio and domain knowledge graph The ratio of the mean of the corresponding ratios Get interactive vocabulary Its profound contributions.

Images