Dialogue chapter parsing method

Abstract

The application discloses a dialogue chapter parsing method, comprising the following steps: obtaining dialogue data; performing text serialization processing on the dialogue data to obtain serialized dialogue history information, chapter relationship category description information and chapter structure annotation information; constructing a chapter parsing model, and inputting the serialized dialogue history information, the chapter relationship category description information and the chapter structure annotation information into the chapter parsing model to train the chapter parsing model; obtaining dialogue information to be parsed, and inputting the dialogue information to be parsed into the trained chapter parsing model to parse the dialogue information to be parsed through the trained chapter parsing model to obtain corresponding chapter structure; thus, the task can be modeled as a text generation task, an additional decoder is avoided, and prediction is completed only through a pre-training model, so that the model performance can be effectively improved by expanding the model size.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and in particular to a method for parsing dialogue texts, a computer-readable storage medium, and a computer device. Background Technology

[0002] In related technologies, multi-person dialogue passage parsing aims to generate a parse tree of the dialogue passage structure, thereby assisting various downstream dialogue tasks, such as dialogue summarization and dialogue response generation. Existing passage parsing models consist of a pre-trained model and a decoder with randomly initialized parameters. Since the additional decoder requires a sufficient amount of data for training, under the realistic condition of limited training data, expanding the scale of the passage parsing model cannot bring about performance improvement. Summary of the Invention

[0003] This invention aims to at least partially solve one of the technical problems in the aforementioned technologies. To this end, one objective of this invention is to propose a dialogue passage parsing method that can model the task as a text generation task, avoiding the introduction of additional decoders and completing predictions solely through a pre-trained model, thereby effectively improving model performance by expanding the model size.

[0004] A second objective of this invention is to provide a computer-readable storage medium.

[0005] The third objective of this invention is to provide a computer device.

[0006] To achieve the above objectives, a first aspect of the present invention proposes a dialogue passage parsing method, comprising the following steps: acquiring dialogue data; performing text serialization processing on the dialogue data to obtain serialized dialogue history information, passage relationship category description information, and passage structure annotation information; constructing a passage parsing model, and inputting the serialized dialogue history information, passage relationship category description information, and passage structure annotation information into the passage parsing model for training; acquiring dialogue information to be parsed, and inputting the dialogue information to be parsed into the trained passage parsing model for parsing the dialogue information to be parsed through the trained passage parsing model to obtain the corresponding passage structure.

[0007] According to the dialogue passage parsing method of the present invention, firstly, dialogue data is acquired; then, the dialogue data is serialized to obtain serialized dialogue history information, passage relationship category description information, and passage structure annotation information; next, a passage parsing model is constructed, and the serialized dialogue history information, passage relationship category description information, and passage structure annotation information are input into the passage parsing model for training; finally, the dialogue information to be parsed is acquired, and the dialogue information to be parsed is input into the trained passage parsing model so that the trained passage parsing model can parse the dialogue information to be parsed to obtain the corresponding passage structure; thus, the task can be modeled as a text generation task, avoiding the introduction of additional decoders, and completing the prediction only through the pre-trained model, thereby effectively improving the model performance by expanding the model size.

[0008] In addition, the dialogue parsing method proposed in the above embodiments of the present invention may also have the following additional technical features:

[0009] Optionally, after obtaining the serialized dialogue history information, text relationship category description information, and text structure annotation information, the serialized dialogue history information, text relationship category description information, and text structure annotation information are further processed by word segmentation and case unification.

[0010] Optionally, the serialized dialogue history information is represented as follows:

[0011] T1,S1:u1T2,S2:u2…T N ,S N :u N

[0012] Among them, t N This indicates that the current round is round N; S N Indicates the name or number of the person speaking in the Nth round; u N This indicates the dialogue history for round N.

[0013] Optionally, the serialized chapter relationship category description information includes sequentially concatenated relationship categories and corresponding descriptions.

[0014] Optionally, the serialized chapter structure annotation information is represented as follows:

[0015] T i ,T j :r ij

[0016] Among them, T i u represents the i-th round of dialogue. i The corresponding identifier; T j u represents the dialogue in round j. j The corresponding identifier; rij representing u i and u j The discourse relation category therebetween; j < i, where i is the current turn.

[0017] Optionally, the serialized discourse structure annotation information includes first structure information, and the first structure information only predicts the discourse relations involved in the current turn.

[0018] Optionally, the serialized discourse structure annotation information includes second structure information, and the second structure information not only needs to predict the discourse relations involved in the current turn, but also needs to predict the discourse relations involved in historical turns.

[0019] Optionally, when parsing the to-be-parsed dialogue information through the trained discourse parsing model, in a constrained decoding manner, predict a discourse structure that meets the task constraints.

[0020] To achieve the above object, a second aspect embodiment of the present invention proposes a computer-readable storage medium, on which a dialogue discourse parsing program is stored, and when the dialogue discourse parsing program is executed by a processor, the dialogue discourse parsing method as described above is implemented.

[0021] According to the computer-readable storage medium of the embodiment of the present invention, by storing the dialogue discourse parsing program, when the processor executes the dialogue discourse parsing program, the dialogue discourse parsing method as described above is implemented. Thus, this task can be modeled as a text generation task, avoiding introducing an additional decoder, and only completing the prediction through a pre-trained model, so that the model performance can be effectively improved by expanding the model scale.

[0022] To achieve the above object, a third aspect embodiment of the present invention proposes a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the dialogue discourse parsing method as described above is implemented.

[0023] According to the computer device of the embodiment of the present invention, by storing the dialogue discourse parsing program in the memory, when the processor executes the dialogue discourse parsing program, the dialogue discourse parsing method as described above is implemented. Thus, this task can be modeled as a text generation task, avoiding introducing an additional decoder, and only completing the prediction through a pre-trained model, so that the model performance can be effectively improved by expanding the model scale. Description of the Drawings

[0024] Figure 1 It is a flowchart of the dialogue discourse parsing method according to the embodiment of the present invention;

[0025] Figure 2 It is a schematic diagram of the overall network structure according to the embodiment of the present invention. Detailed Implementation

[0026] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0027] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the invention to those skilled in the art.

[0028] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0029] Figure 1 This is a flowchart illustrating the dialogue passage parsing method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method for parsing this dialogue passage includes the following steps:

[0030] S101, Obtain dialogue data.

[0031] It should be noted that the dialogue data can be dialogue samples with text structure annotations.

[0032] S102, perform text serialization processing on the dialogue data to obtain serialized dialogue history information, text relationship category description information, and text structure annotation information.

[0033] As an example, after obtaining the serialized dialogue history information, text relationship category description information, and text structure annotation information, word segmentation and case-consistency processing are also performed on the serialized dialogue history information, text relationship category description information, and text structure annotation information.

[0034] As an example, the serialized dialogue history information is represented as follows:

[0035] T1,S1:u1T2,S2:u2…T N ,S N :u N

[0036] Among them, T N This indicates that the current round is round N; S N Indicates the name or number of the person speaking in the Nth round; u NRepresents the dialogue history content of the Nth round.

[0037] As an example, the serialized discourse relation category description information includes the relation category and the corresponding description concatenated in sequence.

[0038] As an example, the serialized discourse structure annotation information is represented as:

[0039] T i ,T j :r ij

[0040] Where, T i represents the identifier corresponding to the dialogue u in the ith round; T i represents the identifier corresponding to the dialogue u in the jth round; r j represents the discourse relation category between u j and u ij ; j < i, that is, i is the current round.

[0041] Where, there can be multiple serialization methods for the discourse structure annotation information.

[0042] As an example, the serialized discourse structure annotation information includes the first structure information, and the first structure information only predicts the discourse relations involved in the current round.

[0043] As another example, the serialized discourse structure annotation information includes the second structure information, and the second structure information not only needs to predict the discourse relations involved in the current round, but also needs to predict the discourse relations involved in the historical rounds.

[0044] As a specific example, as Figure 2 shown, in the example with only 3 rounds of dialogue, the serialized dialogue history information is obtained by concatenating in sequence the round number (such as T0), the name or number of the dialogue participant (such as william), and the dialogue history content (such as anyone have wheat?); the serialized discourse relation category description information is obtained by concatenating in sequence the relation category (such as qap) and the corresponding description (such as an answer to the question); each round of discourse relation annotation in the serialized discourse structure annotation information is obtained by concatenating the items of the relation triple in sequence. For example, for the relation triple <T2, T0, qap> in the figure, its serialized result is T2, T0:qap; specifically, this method includes two optional serialization methods for discourse relation annotation, namely last-turn (LT) and full-history (FH). The former only predicts the discourse relations involved in the current round, while the latter also needs to predict the discourse relations involved in the historical rounds in addition; as​​​​ Figure 2 As shown, LT (left) only needs to predict T2,T0:qap, while FH (right) needs to predict T1,T0:qap and T2,T0:qap. It can be seen that FH not only needs to predict the relationships involved in T2 (current round), but also the relationships involved in T1 (historical round). Compared with LT, FH can use the prediction results of historical rounds to assist in the prediction of the chapter relationships in the current round, but it also introduces additional noise.

[0045] It should be noted that the descriptive information of textual relationships can often be obtained through expert-written guidelines or from dataset annotation documentation. Furthermore, some textual relationships are closely related to keywords; for example, the textual relationship "explanation" can be inferred from the keyword "because" in the dialogue. Therefore, this method also incorporates keyword information enhancement model understanding. Taking "explanation" as an example, its corresponding complete serialization category description is "explanation = explain why, origami the cause of what happened (because)".

[0046] S103. Construct a text parsing model and input the serialized dialogue history information, text relationship category description information, and text structure annotation information into the text parsing model in order to train the text parsing model.

[0047] It should be noted that, because the pre-trained language model T5 has stronger transfer learning capabilities, the discourse parsing model uses the publicly available pre-trained language model T5 for initialization. It predicts the sequential discourse relation labels Y = y1, y2, ..., y3 by inputting a concatenation of serialized dialogue history information and discourse relation category description information X, using an autoregressive generation method. n :

[0048] y i =T5(X,y ＜i )

[0049] During training, the model is trained using maximum likelihood loss:

[0050] Loss = logP(Y|X,θ),

[0051] Here, θ represents the parameters of the model.

[0052] S104: Obtain the dialogue information to be parsed and input it into the trained text parsing model so that the trained text parsing model can parse the dialogue information to be parsed and obtain the corresponding text structure.

[0053] As an example, when parsing the dialogue information to be parsed using a trained discourse parsing model, the discourse structure that satisfies the task constraints is predicted by using constraint decoding.

[0054] It should be noted that, for Figure 2 The dialogue consists of only 3 rounds. The discourse parsing model can predict two sequential discourse relationships. During prediction, the model is forced to generate discourse relationships that meet certain conditions, such as the predicted relationship category must be valid, i.e., it must belong to the set of relationships given by the task. The prediction results are then directly decoded using rules to obtain the corresponding dialogue discourse parsing tree.

[0055] In summary, the dialogue passage parsing method according to embodiments of the present invention first acquires dialogue data; then, it performs text serialization processing on the dialogue data to obtain serialized dialogue history information, passage relationship category description information, and passage structure annotation information; next, it constructs a passage parsing model and inputs the serialized dialogue history information, passage relationship category description information, and passage structure annotation information into the passage parsing model for training; finally, it acquires the dialogue information to be parsed and inputs it into the trained passage parsing model so that the trained passage parsing model can parse the dialogue information to be parsed to obtain the corresponding passage structure; thus, the task can be modeled as a text generation task, avoiding the introduction of additional decoders, and completing the prediction only through a pre-trained model, thereby effectively improving model performance by expanding the model size.

[0056] To implement the above embodiments, this invention provides a computer-readable storage medium storing a dialogue parsing program, which, when executed by a processor, implements the dialogue parsing method as described above.

[0057] According to embodiments of the present invention, a computer-readable storage medium stores a dialogue passage parsing program so that when a processor executes the dialogue passage parsing program, it implements the dialogue passage parsing method as described above. As a result, the task can be modeled as a text generation task, avoiding the introduction of an additional decoder and completing the prediction only through a pre-trained model. This allows for an effective improvement in model performance by expanding the model size.

[0058] To implement the above embodiments, this invention proposes a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the dialogue parsing method described above.

[0059] According to the computer device of the present invention, a dialogue passage parsing program is stored in a memory so that when the processor executes the dialogue passage parsing program, the dialogue passage parsing method described above is implemented. Thus, the task can be modeled as a text generation task, avoiding the introduction of an additional decoder, and the prediction can be completed only through a pre-trained model, thereby effectively improving the model performance by expanding the model size.

[0060] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0061] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0062] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0063] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0064] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0065] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0066] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0067] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0068] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0069] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0070] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0071] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A dialogue discourse parsing method, characterized by, The method comprises the following steps: obtaining dialogue data; performing text serialization processing on the dialogue data to obtain serialized dialogue history information, chapter relationship category description information and chapter structure annotation information; constructing a chapter analysis model and inputting the serialized dialogue history information, chapter relationship category description information and chapter structure annotation information into the chapter analysis model to train the chapter analysis model; obtaining dialogue information to be analyzed and inputting the dialogue information to be analyzed into the trained chapter analysis model to analyze the dialogue information to be analyzed by the trained chapter analysis model to obtain corresponding chapter structure; the serialized dialogue history information is represented as: T1, S1: u1 T2, S2: u2... T N , S N : u N wherein, T N represents the current round is the Nth round; S N represents the name or number of the dialogue person in the Nth round; u N represents the dialogue history content in the Nth round; the serialized discourse relation category description information includes the sequentially spliced relation category and the corresponding description; and the serialized discourse structure annotation information is represented as: T i ,T j :r ij Among them, T i represents the identifier corresponding to the conversation u in the i-th round; T i represents the identifier corresponding to the conversation u in the j-th round; r j represents the identifier corresponding to the conversation u in the j-th round; r j represents the discourse relation category between u ij and u i ; j < i, that is, i is the current round; j ​ when analyzing the dialogue information to be analyzed by the trained chapter analysis model, the chapter structure satisfying the task constraint is predicted in a constraint decoding manner.

2. The dialogue discourse parsing method of claim 1, wherein, After obtaining the serialized dialogue history information, chapter relationship category description information and chapter structure annotation information, the serialized dialogue history information, chapter relationship category description information and chapter structure annotation information are also subjected to word segmentation and case unification processing.

3. The discourse parsing method of claim 2, wherein, The serialized chapter structure annotation information comprises first structure information, and the first structure information only predicts the chapter relationship involved in the current round.

4. The dialogue discourse parsing method of claim 2, wherein, The serialized chapter structure annotation information comprises second structure information, and the second structure information not only needs to predict the chapter relationship involved in the current round, but also needs to predict the chapter relationship involved in the historical round.

5. A computer readable storage medium, characterized in that, A storage medium having stored thereon a dialogue chapter analysis program which, when executed by a processor, implements the dialogue chapter analysis method according to any one of claims 1-4.

6. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the dialogue chapter analysis method according to any one of claims 1-4 when executing the program.

Images