A structured dynamic semantic reconstruction method, device and storage medium

Abstract

The embodiment of the application discloses a structured dynamic semantic reconstruction method, device and medium. The method comprises: in the process of dynamic dialogue reconstruction, obtaining to-be-processed data; inputting the to-be-processed data into a pre-trained dynamic semantic reconstruction model, and outputting a result of semantic reconstruction of speaking information of a current last person. The application enhances the understanding ability of the model to the sentence with complex syntax structure by performing syntax dependency analysis on the sentence and fusing the syntax dependency analysis into the word coding, position coding and other modules of the model, so that the sentence can be correctly and smoothly reconstructed, and the semantic understanding deviation problem of the algorithm model caused by the existence of reference and omission in human-computer dynamic interaction is solved. Meanwhile, the application does not need a large amount of manual annotation work, and can be trained only by using dialogue data, thereby reducing the dependence on high-quality manual annotation data in the training process of the algorithm model, and reducing the dependence on hardware materials required for deployment of the algorithm model.

Description

A structured dynamic semantic reconstruction method, apparatus, and storage medium Technical Field

[0001] This invention relates to the field of artificial intelligence question answering technology, specifically to a structured dynamic semantic reconstruction method, apparatus, and storage medium. Background Technology

[0002] Dynamic dialogue semantic reconstruction is a method designed to address semantic comprehension biases in real-world human-computer interaction using artificial intelligence. Semantic comprehension biases in human-computer interaction typically arise during dialogues between humans and algorithmic models. In everyday conversations, because the human brain records historical information about the dialogue, human speech often involves referencing and omitting to reduce redundancy. Referencing refers to using demonstrative pronouns to refer to a fixed or abstract concept, such as A: I like watching *Chinese Paladin 3*. B: I also like watching this. Here, "this" refers to *Chinese Paladin 3* mentioned earlier. Omission refers to simplifying the response without affecting understanding; it's a special type of referencing, also known as zero referencing, for example, A: I like watching *Chinese Paladin 3*. B: Me too. Clearly, B means that I also like *Chinese Paladin 3*.

[0003] The human brain possesses memory capabilities and can effectively reconstruct important information from the history of a conversation. However, in algorithmic models, accurately understanding a user's semantics typically requires the user to provide complete and detailed information, which differs significantly from human behavioral habits. Therefore, achieving semantic reconstruction of dynamic dialogue through artificial intelligence algorithmic models is particularly important.

[0004] The existing technologies and their characteristics are as follows:

[0005] Existing technology 1: Intelligent customer service technology based on dialogue state tracking.

[0006] This technology is commonly found in various applications, such as food ordering, ticket booking, weather inquiries, logistics tracking, and product consultations. First, it uses intent recognition algorithms to detect the user's true intent, such as when booking tickets. Second, it uses slot-filling methods to complete the service flow with key information; for example, booking tickets requires knowing the departure point, destination, and departure time. Finally, it uses dialogue state tracking and management to repeatedly fill the slots, providing semantically clear and explicit user requirements for subsequent tasks. However, this method is complex, inflexible, and requires pre-creating a large number of slot templates. As the application scenario becomes more complex, the number of slot templates increases accordingly, gradually becoming redundant, complex, and difficult to maintain.

[0007] Existing technology 2: Sequence labeling algorithm based on deep learning.

[0008] In the field of Natural Language Processing, there is a task called sequence labeling, which extracts key information by labeling text with fixed tags. In the example "A: I like 'The Legend of Sword and Fairy 3'. B: I like it too," "'The Legend of Sword and Fairy 3'" would be labeled as a specific tag. By concatenating multiple similar key-specific fragments, semantics are completed. The disadvantages of this method are that it firstly requires a large amount of manual labeling for training the algorithm model; secondly, it lacks flexibility, as the manual labeling task needs to be repeated when the application scenario changes, resulting in poor scalability and high cost; finally, the semantics obtained through the concatenation strategy are prone to incoherence.

[0009] Existing technology three: a method based on modern large models.

[0010] With the advancements in capabilities of large-scale models like Tongyi Qianwen and Zhipu Qingyan, a significant amount of research has utilized these models for semantic refinement. Such methods are effective, easily scalable, and capable of dynamically understanding contextual semantic information. However, their drawbacks are also apparent: firstly, large-scale model-based methods rely on high-quality prompt word templates; secondly, the reasoning speed of large-scale models is relatively slow, as semantic refinement typically occurs within a single stage of the dialogue flow, leading to frequent calls to the large-scale model during dialogues; and finally, the effectiveness of large-scale models is closely related to their massive parameter set, making high-quality hardware a major challenge for deployment.

[0011] Each of the above methods has its own advantages and disadvantages in solving practical problems. In reality, high-quality manual annotation and high-quality hardware materials are difficult to obtain and are very expensive. Summary of the Invention

[0012] The purpose of this invention is to provide a structured dynamic semantic reconstruction method, apparatus, and storage medium to solve the problem of semantic understanding deviation in algorithm models caused by the existence of referential and omission phenomena in human-computer dynamic interaction. At the same time, this invention can also reduce the need for high-quality manually labeled data for algorithm model training and reduce the dependence on hardware materials required for algorithm model deployment.

[0013] To achieve the above objectives, in a first aspect, embodiments of the present invention provide a structured dynamic semantic reconstruction method, comprising:

[0014] During the dynamic dialogue reconstruction process, data to be processed is acquired; the data to be processed includes historical dialogue information and the speech information of the last person currently speaking.

[0015] The data to be processed is input into a pre-trained dynamic semantic reconstruction model for processing, and the output is the semantic reconstruction result of the speech information of the last person.

[0016] The dynamic semantic reconstruction model includes:

[0017] The semantic structure encoding module is used to perform syntactic dependency analysis on the input text, identify the dependency relationships between the components within the sentence, and convert the dependency relationships into vector representations.

[0018] The encoding enhancement representation module, based on a pre-trained Transformer encoder architecture, is used for:

[0019] The semantic results processed by the semantic structure encoding module are encoded using an attention mechanism and improved location information encoding technology, and syntactic cues are used to help the dynamic semantic structure model interpret user intent; and

[0020] The semantic content decoder, based on a pre-trained Transformer decoder structure, is used to decode the encoding results obtained by the encoding enhancement representation module.

[0021] Furthermore, as a preferred implementation of this application, the method further includes constructing a dynamic semantic reconstruction model before acquiring the data to be processed.

[0022] As a specific implementation of this application, the data to be processed is input into a pre-trained dynamic semantic reconstruction model for processing, specifically as follows:

[0023] Syntactic dependency analysis is performed on the data to be processed to identify the dependency relationships between the components within the sentence, thereby obtaining the syntactic dependency embedding results. The dependency relationships are then converted into vector representations to obtain the semantic embedding results.

[0024] The semantic embedding results and the syntactic dependency embedding results are fused to obtain word representations;

[0025] The word representations are input into the Transformer encoder to generate context-aware representations;

[0026] The word representation and context-aware representation are transformed into vectors to obtain a sentence encoding vector with syntactic distance;

[0027] The sentence encoding vector is input into the Transformer decoder, which outputs a semantically fluent sentence.

[0028] Secondly, embodiments of this application also provide a structured dynamic semantic reconstruction apparatus, comprising:

[0029] The data acquisition unit is used to acquire data to be processed during the dynamic dialogue reconstruction process; the data to be processed includes historical dialogue information and the speech information of the last person currently speaking.

[0030] The semantic reconstruction unit is used to input the data to be processed into a pre-trained dynamic semantic reconstruction model for processing, and output the semantic reconstruction result of the speech information of the last person.

[0031] The dynamic semantic reconstruction model includes a semantic structure encoding module, an encoding enhancement representation module, and a semantic content decoder.

[0032] Thirdly, embodiments of the present invention also provide a structured dynamic semantic reconstruction apparatus, including a processor, an input device, an output device, and a memory, wherein the processor, input device, output device, and memory are interconnected, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to invoke the program instructions to execute the method described in the first aspect.

[0033] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, cause the processor to perform the method described in the first aspect.

[0034] By implementing embodiments of the present invention, syntactic dependency analysis is performed on sentences and integrated into the word encoding, position encoding, and other modules of the model. This enhances the model's ability to understand sentences with complex syntactic structures, thereby ensuring correct and fluent sentence reconstruction. This solves the problem of semantic understanding deviation in algorithm models caused by the existence of pronouns and omissions in human-computer dynamic interaction. At the same time, the present invention does not require a large amount of manual annotation work; it can be trained using only dialogue data. This reduces the dependence of the algorithm model on high-quality manually annotated data during training and also reduces the dependence on hardware materials required for the deployment of the algorithm model. Attached Figure Description

[0035] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below.

[0036] Figure 1 is a flowchart of the structured dynamic semantic reconstruction method provided in an embodiment of the present invention;

[0037] Figure 2 is a block diagram of the dynamic semantic reconstruction model;

[0038] Figure 3 is another structural diagram of the dynamic semantic reconstruction model;

[0039] Figure 4 is a structural diagram of the structured dynamic semantic reconstruction device provided in an embodiment of the present invention;

[0040] Figure 5 is another structural diagram of the device shown in Figure 4. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0043] Please refer to Figure 1, which illustrates a structured dynamic semantic reconstruction method provided in an embodiment of the present invention, which may include the following steps:

[0044] S1, Construct a dynamic semantic reconstruction model.

[0045] Please refer to Figures 2 and 3. This dynamic semantic reconstruction model includes a semantic structure encoding module, an encoding enhancement representation module, and a semantic content decoder.

[0046] The semantic structure encoding module is used to perform syntactic dependency analysis on the input text, identify the dependency relationships between the components within the sentence, and convert the dependency relationships into vector representations. This module explicitly injects the structural information of the text into the model.

[0047] The encoding enhancement representation module, based on a pre-trained Transformer encoder structure, employs bidirectional encoding combined with an autoregressive generation strategy; this module is used for:

[0048] The semantic results processed by the semantic structure encoding module are encoded using an attention mechanism and improved location information encoding technology, and syntactic cues are used to help the dynamic semantic structure model interpret user intent; and

[0049] The semantic content decoder, based on a pre-trained Transformer decoder structure, is used to decode the encoded result obtained by the encoding enhancement representation module. That is, this module primarily generates the final output from the semantic representation passed from the encoder. It progressively generates the target sequence using previously acquired structural and semantic information.

[0050] The following is a detailed explanation of several modules in the model:

[0051] (I) Semantic Structure Encoding Module

[0052] In the process of dynamic dialogue reconstruction, capturing the complex relationships between dialogues is crucial for obtaining coherent semantic information. Dependency parsing techniques can capture the grammatical relationships between words in a sentence and provide a structured representation. The semantic structure encoding module then uses vector encoding of the results of the dependency parsing tree, allowing it to be flexibly combined with the semantic content module. The following will use the example of "A: I like Chinese Paladin 3, B: I like it too" to illustrate the modeling method of this module in detail:

[0053] a) First, the mathematical representation of this module is described.

[0054] For an input consisting of multiple sentences, each sentence is segmented, and syntactic dependency parsing is used to obtain the syntactic structure. For a sentence U=[w i w2, ..., w n The dependency parser generates a set of dependency arcs:

[0055]

[0056] in, It is the central word. It is a dependency word, r i It is a relation type (e.g., subject, predicate, object).

[0057] For each word w i We will construct a syntactic embedding S containing its dependencies. i :

[0058]

[0059] in Indicates the connection between the dependency arc and the word w i Connected neighboring word sets, f dep It is a function that aggregates syntactic information. We define f dep for:

[0060]

[0061] Where e j and e k They are respectively with w i Connected dependency words and headword embeddings, W r It is a relation-specific weight matrix, and the summation operation iterates through all the weights related to w. i The relevant outgoing and incoming arcs.

[0062] b) The above process will now be described explanatoryly.

[0063] First, construct the syntactic dependency tree.

[0064] I like Chinese Paladin 3. I am also the root node. I like leaf node 1. I [subject] I [subject] Leaf node 2. Chinese Paladin 3 [direct object] also [adverbial] surface

[0065] Table 1: Syntactic Dependency Tree

[0066] Next, syntactic embeddings are constructed. For each word in each sentence, syntactic embeddings are constructed based on their dependency relations. Sentence 1: Words: I (w1), like (w2), Legend of Sword and Fairy 3 (w3). Dependency relations are (w2, w1, subject)(w2, w3, direct object). i Syntactic embeddings i The calculation formula is as described in formula (3). For the word w1 (I), its dependency relation (w2, w1, subject) is passed in, then s1 = W nsubj e2, where nsubj represents the "subject". Similarly, the calculation method for other words and sentences is the same. At this point, the syntactic dependency embedding of each sentence is complete.

[0067] (II) Encoding Enhancement Representation Module

[0068] In the process of dynamic dialogue reconstruction, the model not only needs to understand the semantic information of each dialogue, but also needs to capture the syntactic structure to ensure that the output content is grammatically correct and coherent. To this end, this invention proposes to fuse semantic content embedding and syntactic dependency embedding, and to handle dependency relationships through a weight matrix. The specific method is described as follows:

[0069] The first step involves fusing semantic content embeddings and syntactic dependency embeddings. Semantic content embeddings are typically used to represent the semantic information of words. A common approach is to use pre-trained language models such as Word2Vector and BERT to encode each word into a high-dimensional dense vector. We represent the semantic embedding of each semantically encoded word w1 as e1, and fuse the syntactic dependency embedding s1 mentioned above through a linear transformation. The model automatically optimizes the fusion parameters during training.

[0070] h i =tanh(W c [e i ;s i ]+b c (4)

[0071] Where [·;·] denotes the vector concatenation operation, W c and b cThese are trainable parameters, and tanh is a non-linear activation function. This mechanism combines semantic and syntactic information to generate word representations h that contain richer information. i .

[0072] The second step is to express the word h. i The input is fed into the encoder part of the Transformer to generate a context-aware representation Z. i Encoders typically consist of multiple layers, each including a self-attention mechanism and a feedforward neural network. Through the stacking of multiple layers, the model can capture long-distance dependencies between words. Transformer-based encoder models have become a general structure in the field of natural language processing and are not the core work of this invention; therefore, their principles will not be elaborated upon here. i The calculation process is as follows:

[0073] z i =TransformerEncoderLayer(h i (5)

[0074] TransformeEncodeLayer represents the encoder described above.

[0075] The third step, in constructing syntactic dependency embeddings, typically generates a syntactic dependency tree, as shown in Table 1. The dependency tree reflects the grammatical hierarchy between words in a sentence; words with shallower depths (closer to the root node) play a more central role. For example, subjects and predicates are usually located at shallower levels. Deeper words are often modifiers or clause components, serving to supplement or modify the sentence. Furthermore, the depth of the tree also reflects the complexity of the sentence. Greater depth indicates a more complex sentence structure with more dependencies such as modifiers and clauses, often allowing the sentence to express richer information.

[0076] We fuse the depth information of the dependency tree with the positional information encoding of the model to form a positional information encoding with syntactic distance. Since the Transformer model itself does not have the ability to process sequence order, a positional information encoding module is usually added to it. A common approach is to use sine and cosine functions to generate the encoding PE(w) for each word position. i The specific principles of positional encoding are beyond the scope of this invention and will not be elaborated upon here. We will integrate them with syntactic depth.

[0077] de i =PE(w i )+λ·SD(w i (6)

[0078] Here, λ is an adjustable scaling factor used to control the degree of influence of syntactic distance on the final positional encoding. By adjusting λ, the influence of position and syntactic structure on positional encoding can be balanced. SD(w i ) indicates the word w i The depth in the syntactic dependency tree reflects its hierarchy in the syntactic structure. The final positional information encoding with syntactic distance is obtained through formula (6). i .

[0079] At this point, we will de i With s i e i The two vectors are added together, and the final sentence encoding vector is obtained by calculation using formulas (4) and (5).

[0080] (III) Semantic Content Decoder

[0081] The semantic content decoding module consists of the decoder part of the Transformers model, which receives the aforementioned encoded vectors as input and outputs semantically fluent sentences. Architecturally, it incorporates self-attention mechanisms, cross-attention mechanisms, and feedforward neural networks.

[0082]

[0083] in It is the output of the l-th layer decoder. H is the initial input embedding vector, which is usually a randomly initialized vector, and H is the comprehensive semantic representation output by the encoder.

[0084] S2 acquires the data to be processed during the dynamic dialogue reconstruction process.

[0085] The data to be processed includes historical dialogue information and the speech information of the last person currently speaking.

[0086] S3, input the data to be processed into the pre-trained dynamic semantic reconstruction model for processing, and output the semantic reconstruction result of the speech information of the last person.

[0087] In specific implementation, the data to be processed is input into a pre-trained dynamic semantic reconstruction model for processing, specifically as follows:

[0088] Syntactic dependency analysis is performed on the data to be processed to identify the dependency relationships between the components within the sentence, thereby obtaining the syntactic dependency embedding results. The dependency relationships are then converted into vector representations to obtain the semantic embedding results.

[0089] The semantic embedding results and the syntactic dependency embedding results are fused to obtain word representations;

[0090] The word representations are input into the Transformer encoder to generate context-aware representations;

[0091] The word representation and context-aware representation are transformed into vectors to obtain a sentence encoding vector with syntactic distance;

[0092] The sentence encoding vector is input into the Transformer decoder, which outputs a semantically fluent sentence.

[0093] It should be noted that for a more detailed description of step S3, please refer to the aforementioned detailed description of each module in the model.

[0094] The structured dynamic semantic reconstruction method provided in this invention enhances the model's ability to understand sentences with complex syntactic structures by performing syntactic dependency analysis on sentences and integrating it into the model's word encoding and position encoding modules. This ensures correct and fluent sentence reconstruction and solves the problem of semantic understanding deviation in algorithm models caused by the existence of pronouns and omissions in human-computer dynamic interaction. At the same time, this invention does not require a large amount of manual annotation work and can be trained using only dialogue data, reducing the dependence of the algorithm model on high-quality manually annotated data during training and reducing the dependence on hardware materials required for the deployment of the algorithm model.

[0095] Based on the same inventive concept, as shown in Figure 4, this embodiment of the invention also provides a structured dynamic semantic reconstruction device, comprising:

[0096] A model building unit is used to build a dynamic semantic reconstruction model; wherein, the dynamic semantic reconstruction model includes a semantic structure encoding module, an encoding enhancement representation module, and a semantic content decoder;

[0097] The data acquisition unit is used to acquire data to be processed during the dynamic dialogue reconstruction process; the data to be processed includes historical dialogue information and the speech information of the last person currently speaking.

[0098] The semantic reconstruction unit is used to input the data to be processed into a pre-trained dynamic semantic reconstruction model for processing, and output the semantic reconstruction result of the speech information of the last person.

[0099] The semantic structure encoding module is used to perform syntactic dependency analysis on the input text, identify the dependency relationships between the components within the sentence, and convert the dependency relationships into vector representations.

[0100] The encoding enhancement representation module, based on a pre-trained Transformer encoder structure, encodes the semantic results processed by the semantic structure encoding module through an attention mechanism and improved positional information encoding technology, and uses syntactic cues to help the dynamic semantic structure model interpret user intent.

[0101] The semantic content decoder, based on a pre-trained Transformer decoder structure, is used to decode the encoding results obtained by the encoding enhancement representation module.

[0102] Furthermore, the semantic reconstruction unit is specifically used for:

[0103] Syntactic dependency analysis is performed on the data to be processed to identify the dependency relationships between the components within the sentence, thereby obtaining the syntactic dependency embedding results. The dependency relationships are then converted into vector representations to obtain the semantic embedding results.

[0104] The semantic embedding results and the syntactic dependency embedding results are fused to obtain word representations;

[0105] The word representations are input into the Transformer encoder to generate context-aware representations;

[0106] The word representation and context-aware representation are transformed into vectors to obtain a sentence encoding vector with syntactic distance;

[0107] The sentence encoding vector is input into the Transformer decoder, which outputs a semantically fluent sentence.

[0108] It should be noted that the specific workflow of this embodiment is described in the foregoing method embodiment section, and will not be repeated here.

[0109] Furthermore, another embodiment of the present invention provides a structured dynamic semantic reconstruction apparatus. As shown in FIG5, the apparatus may include: one or more processors 101, one or more input devices 102, one or more output devices 103, and a memory 104, wherein the processors 101, input devices 102, output devices 103, and memory 104 are interconnected via a bus 105. The memory 104 is used to store a computer program, the computer program including program instructions, and the processor 101 is configured to invoke the program instructions to execute the method described in the above-described method embodiment.

[0110] It should be understood that, in this embodiment of the invention, the processor 101 may be a central processing unit (CPU), but it may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0111] Input device 102 may include a keyboard, etc., and output device 103 may include a display (LCD, etc.), a speaker, etc.

[0112] The memory 104 may include read-only memory and random access memory, and provides instructions and data to the processor 101. A portion of the memory 104 may also include non-volatile random access memory. For example, the memory 104 may also store device type information.

[0113] In specific implementations, the processor 101, input device 102, and output device 103 described in the embodiments of the present invention can execute the implementation methods described in the embodiments of the structured dynamic semantic reconstruction method provided in the embodiments of the present invention, which will not be repeated here.

[0114] Accordingly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, implement the above-described structured dynamic semantic reconstruction method.

[0115] The computer-readable storage medium can be an internal storage unit of the system described in any of the foregoing embodiments, such as the system's hard disk or memory. The computer-readable storage medium can also be an external storage device of the system, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the computer-readable storage medium can include both internal storage units and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the system. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0116] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0117] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices or units, or may be electrical, mechanical or other forms of connection.

[0118] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0119] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0120] If the integrated unit is implemented as 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 technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0121] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A structured dynamic semantic reconstruction method, characterized in that, include: During the dynamic dialogue reconstruction process, data to be processed is acquired; The data to be processed includes historical dialogue information and the speech information of the last person currently speaking. The data to be processed is input into a pre-trained dynamic semantic reconstruction model for processing, and the output is the semantic reconstruction result of the speech information of the last person. The dynamic semantic reconstruction model includes a semantic structure encoding module, an encoding enhancement representation module, and a semantic content decoder. Before acquiring the data to be processed, the method further includes constructing the dynamic semantic reconstruction model. Specifically, the data to be processed is input into the pre-trained dynamic semantic reconstruction model for processing: syntactic dependency analysis is performed on the data to be processed to identify the dependency relationships between the components within the sentence, obtaining syntactic dependency embedding results, and the dependency relationships are converted into vector representations to obtain semantic embedding results. During the construction of syntactic dependency embeddings, a syntactic dependency tree is generated, which reflects the grammatical hierarchy between words in the sentence. The semantic embedding results and the syntactic dependency embedding results are fused to obtain a vocabulary list. The process involves inputting the word representations into a Transformer encoder to generate context-aware representations; performing vector transformation on the word representations and context-aware representations to obtain sentence encoding vectors with syntactic distance; inputting the sentence encoding vectors into a Transformer decoder to output semantically fluent sentences; and using the encoding enhancement module, based on a pre-trained Transformer encoder structure, to: encode the semantic results processed by the semantic structure encoding module through an attention mechanism and improved positional information encoding techniques, using syntactic cues to help the dynamic semantic structure model interpret user intent; wherein, the improved positional information encoding techniques include: fusing the depth information of the dependency tree with the model's positional information encoding to form positional information encoding with syntactic distance, specifically including: using sine and cosine functions to generate the encoding for each word position. Encoding syntactic depth and word position Integration; ,in, It is an adjustable scaling factor used to control the degree to which syntactic distance affects positional encoding. This scaling factor is adjusted... The impact of balancing position and syntactic structure on positional encoding Words Depth in the syntactic dependency tree This indicates the location information encoding.

2. The structured dynamic semantic reconstruction method as described in claim 1, characterized in that, The semantic structure encoding module is used to: perform syntactic dependency analysis on the input text, identify the dependency relationships between the components within the sentence, and convert the dependency relationships into vector representations.

3. The structured dynamic semantic reconstruction method as described in claim 2, characterized in that, The semantic content decoder is based on a pre-trained Transformer decoder structure and is used to decode the encoding result obtained by the encoding enhancement representation module.

4. A structured dynamic semantic reconstruction device, characterized in that, The system includes a processor, an input device, an output device, and a memory, which are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to invoke the program instructions to perform the method as described in any one of claims 1-3.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method as described in any one of claims 1-3.

Images