CN117131187B - Dialogue abstracting method based on noise binding diffusion model - Google Patents
Dialogue abstracting method based on noise binding diffusion model Download PDFInfo
- Publication number
- CN117131187B CN117131187B CN202311395915.2A CN202311395915A CN117131187B CN 117131187 B CN117131187 B CN 117131187B CN 202311395915 A CN202311395915 A CN 202311395915A CN 117131187 B CN117131187 B CN 117131187B
- Authority
- CN
- China
- Prior art keywords
- noise
- dialogue
- diffusion
- abstract
- binding
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000009792 diffusion process Methods 0.000 title claims abstract description 82
- 238000000034 method Methods 0.000 title claims abstract description 76
- 238000012512 characterization method Methods 0.000 claims abstract description 29
- 230000008447 perception Effects 0.000 claims abstract description 11
- 239000013598 vector Substances 0.000 claims description 24
- 230000002452 interceptive effect Effects 0.000 claims description 11
- 238000005070 sampling Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 6
- 239000000203 mixture Substances 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 3
- 238000012804 iterative process Methods 0.000 claims description 3
- 230000003993 interaction Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000008451 emotion Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/34—Browsing; Visualisation therefor
- G06F16/345—Summarisation for human users
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention relates to the technical field of dialogue abstract generation, and discloses a dialogue abstract method based on a noise binding diffusion model, wherein the training process comprises the following steps: generating corresponding inquiry information according to the input dialogue; inputting the query information and the dialogue in series connection and splicing into a dialogue encoder to obtain a dialogue characterization of query perception; randomly generating noise information, converting the artificially marked dialogue abstract into binary bit representation, and performing diffusion noise adding processing by using the noise information to obtain a noise adding representation; denoising the denoised representation by using a diffusion encoder-decoder to obtain a denoised bit representation, comparing the denoised bit representation with the bit representation to obtain a diffusion loss, and updating model parameters of the diffusion encoder-decoder by using a back propagation algorithm through the diffusion loss; by using the diffusion coder-decoder based on noise binding to perform digest generation of different roles, irrelevant information serving as noise can be effectively distinguished from the dialogue, and the quality of the generated digest is improved.
Description
Technical Field
The invention relates to the technical field of dialog digest generation, in particular to a dialog digest method based on a noise binding diffusion model.
Background
The session summarization task aims at summarizing the expressed core content of different roles from the session. The invention solves the following two technical problems:
existing methods generally use an autoregressive model paradigm to capture key information from a conversation by means of a attentive mechanism, but they cannot distinguish a large amount of irrelevant information (such as greetings, emotion words, etc.) commonly existing in the conversation, and these irrelevant information act as noise in the process of generating a summary, which easily causes the problem that the autoregressive model is spread in error in the process of generating. In this regard, the present invention proposes a non-autoregressive digest generation method that binds noise to irrelevant content, and distinguishes important information from irrelevant information in a conversation, thereby generating a more efficient digest.
Existing methods tend to generate summaries for different roles separately, neglecting interactions between different roles in the conversation, which results in the summaries of different roles having non-matching questions in some scenarios, e.g., user questions involved in user summaries have no corresponding answers in customer service summaries. In this regard, the invention provides an interactive perception noise binding method, which introduces interactive information among different roles when preprocessing noise in data, so that the abstract model can more effectively utilize the interaction among roles to generate an abstract with high correlation.
Disclosure of Invention
In order to solve the technical problems, the invention provides a dialogue abstracting method based on a noise binding diffusion model.
In order to solve the technical problems, the invention adopts the following technical scheme:
dialogue abstracting method based on noise binding diffusion modelInputting a dialogue abstract model to generate a dialogue abstract; the dialogue abstract model comprises a noise binding module, a query generation module, a dialogue encoder and a diffusion encoder-decoder; the diffusion encoder-decoder includes a diffusion encoder and a diffusion decoder;
a training process for a dialog summary model, comprising the steps of:
s1, using a query generation module, according to the input dialogueGenerating corresponding inquiry information->;
S2, inquiring the informationAnd dialogue->Concatenation is carried out in series, and a dialogue encoder is input to obtain a dialogue representation of query perception;
S3, randomly generating noise vectors through a noise binding moduleAnd abstracts the manually marked dialog +.>Bit representation converted into binary>Diffusion encoder uses noise vector +>For->Performing diffusion noise treatment, wherein the diffusion noise treatment process comprises +.>Step, obtaining a noise adding representation obtained in the step t of diffusion noise adding treatment>Noisy characterization for and last step;
S4, characterizing the dialogueInput diffusion encoder-decoder and use diffusion decoder +.>Adding toNoise characterization->Performing denoising treatment, wherein the denoising treatment comprises +.>Step, get the denoising treatment +.>Step De-N ratio characterization->De-N ratio characterization of the last step->Predicted dialog abstract->Denoising treatment->Step De-N ratio characterization->Characterization of the noise obtained in step t ∈ ->Comparison gives diffusion loss->The method comprises the steps of carrying out a first treatment on the surface of the Predicted dialogue abstract->Dialog abstract with manual annotation->In contrast, calculate loss->The method comprises the steps of carrying out a first treatment on the surface of the Total loss->The method comprises the steps of carrying out a first treatment on the surface of the By total loss->And updating model parameters of the diffusion encoder-decoder using a back propagation algorithm; that is, only model parameters of the diffusion encoder-decoder are updated during dialogue digest model training;
in step S3, a noise vector is randomly generated through a noise binding moduleThe method specifically comprises the following steps: generating a noise set by means of a noise binding module, randomly sampling +.>A noise word, using an embedding layer matrix +.>Converting each noise word into a noise representation +.>Then average all the noise characterizations and normalizeObtaining noise vector->;/>Indicate->A noise word;
in step S3, the artificially marked dialogue abstract is summarizedBit representation converted into binary>Using noise vectorsFor->Performing diffusion noise adding treatment to obtain noise adding representation->The method specifically comprises the following steps: summarizing conversational sessions based on a bit-diffusion networkBit representation of the word in (a) converted into binary>Then use the noise vector +.>For->Performing diffusion noise treatment, wherein the diffusion noise treatment process comprises +.>Step (2) diffusion noise treatment the noise characterization obtained in step (t)>The method comprises the following steps:;/>representing a predefined hyper-parameter;
in step S4, the diffusion lossThe method comprises the following steps: />The method comprises the steps of carrying out a first treatment on the surface of the Wherein->For calculating the length of the vector; by means of a loss function->Calculating the loss->:/>The method comprises the steps of carrying out a first treatment on the surface of the Loss functionIs a cross entropy loss function.
Further, the dialogue is to be performedThe specific process of inputting a dialogue abstract model and generating the dialogue abstract comprises the following steps:
s51: based on the input dialog, using a query generation moduleGenerating corresponding inquiry information->;
S52: will beAnd->Concatenation, input dialogue encoder, get query perceived dialogue characterization +.>;
S53: generating a noise set by a noise binding module, randomly sampling from the noise set to obtain a noise vectorWill->Initial state as dialog digest generation procedure +.>Use of a diffusion codec pair +.>Iterative denoising is carried out to obtain binary bit representation corresponding to the dialogue abstract>Will->Converting into decimal system to obtain final dialog abstract +.>。
Further, in step S53, when generating the noise set by the noise binding module, the method specifically includes:
dialog for a conversationThe vocabulary of all word compositions in +.>The vocabulary of all word components in the manually annotated dialog abstract is +.>The words which do not appear in the dialogue abstract are all regarded as noise information and are treated as word list +.>He vocabulary->Collecting difference set operation to obtain universal binding noise set +.>;
Recording the manually marked user inquiry set asWherein->Representing the user query set +.>The M-th individual query of (1) remembers the word list of the word composition in all individual queries as +.>Words which do not appear in the individual queries are regarded as noise information by means of the vocabulary +.>He vocabulary->Collecting difference set operation to obtain interactive perception binding noise set;
The final noise set is obtained by carrying out union operation on the universal binding noise set and the interactive perception binding noise set。
Further, in step S53, a finally generated dialog abstract is obtainedThe specific process of (2) comprises:
random sampling from noise setsEach noise word is converted into a noise representation by using an embedding layer matrix, and then all the noise representations are averaged and normalized to obtain a noise vector +.>Will->Initial state as dialog digest generation procedure +.>;
Diffusion decoderFor->Performing iterative denoising, wherein the iterative process comprises->Step->The characteristic of the denoising ratio obtained after the step denoising is marked as +.>First->Characterization of the noise ratio obtained in the step->;
Wherein the intermediate variable
;
Representing a predefined hyper-parameter;
characterization of the resulting binary bitsConverting into decimal system to obtain the final productGenerated dialog abstract->。
Compared with the prior art, the invention has the beneficial technical effects that:
by using the diffusion coder-decoder based on noise binding to perform digest generation of different roles, irrelevant information serving as noise can be effectively distinguished from the dialogue and removed from the dialogue, so that the generated digest content can capture core content in the dialogue more, and the quality of the generated digest is improved.
Through the interactive perception noise binding based on the query-summary pair, the interactive information between the roles is fused into the noise, so that the model can fully utilize the interactive information between different roles in the summary generation process, further assist the summary generation process, enable the summary content relevance between different generated roles to be higher, further promote the summary generation performance of the model, and enable the model to be more suitable for actual use scenes.
Drawings
Fig. 1 is a flow chart of a method of abstracting a dialogue in the present invention.
Detailed Description
A preferred embodiment of the present invention will be described in detail with reference to the accompanying drawings.
The invention provides a dialogue abstracting method based on a noise binding diffusion model, which is used for dialogingInputting a dialogue abstract model to generate a dialogue abstract; the dialogue abstract model comprises a noise binding module, a query generation module, a dialogue encoder and a diffusion encoder-decoder; the diffusion encoder-decoder includes a diffusion encoder and a diffusion decoder; the whole flow of the dialogue abstracting method is shown in fig. 1, and abstracting generating processes of different roles follow the same flow.
A training process for a dialog summary model, comprising the steps of:
s1, using a query generation module to generate a query according to the inputIncoming conversationsGenerating corresponding inquiry information->。
S2, inquiring the informationAnd dialogue->Concatenation is carried out in series, and a dialogue encoder is input to obtain a dialogue representation of query perception。
S3, randomly generating noise vectors through a noise binding moduleAnd abstracts the manually marked dialog +.>Bit representation converted into binary>Diffusion encoder uses noise vector +>For->Performing diffusion noise treatment, wherein the diffusion noise treatment process comprises +.>Step, obtaining a noise adding representation obtained in the step t of diffusion noise adding treatment>Noisy characterization for and last step。
In step S3, a noise vector is randomly generated through a noise binding moduleThe method specifically comprises the following steps: generating a noise set by means of a noise binding module, randomly sampling +.>A noise word, using an embedding layer matrix +.>Converting each noise word into a noise representation +.>Then average all the noise characterizations and normalizeObtaining noise vector->;/>Indicate->A noise word.
In step S3, the artificially marked dialogue abstract is summarizedBit representation converted into binary>Using noise vectorsFor->Performing diffusion noise adding processing to obtain a noise adding representation/>The method specifically comprises the following steps: summarizing conversational sessions based on a bit-diffusion networkBit representation of the word in (a) converted into binary>Then use the noise vector +.>For->Performing diffusion noise treatment, wherein the diffusion noise treatment process comprises +.>Step (2) diffusion noise treatment the noise characterization obtained in step (t)>:;/>Representing predefined hyper-parameters. The bit diffusion network adopted by the invention is the prior technical scheme, and can be specifically referred to in the literature of Analog Bits Generating Discrete Data using Diffusion Models with Self-Conditioning.
S4, characterizing the dialogueInput diffusion encoder-decoder and use diffusion decoder +.>Characterization of noise->Performing denoising treatmentThe denoising process also includes->Step, get the denoising treatment +.>Step De-N ratio characterization->De-N ratio characterization of the last step->Predicted dialog abstract->Denoising treatment->Step De-N ratio characterization->Characterization of the noise obtained in step t ∈ ->Comparison gives diffusion loss->The method comprises the steps of carrying out a first treatment on the surface of the Predicted dialogue abstract->Dialog abstract with manual annotation->In contrast, calculate loss->The method comprises the steps of carrying out a first treatment on the surface of the Total loss->The method comprises the steps of carrying out a first treatment on the surface of the By total loss->And updating model parameters of the diffusion encoder-decoder using a back propagation algorithm; that is, only model parameters of the diffusion codec are updated at the time of dialogue digest model training.
In step S4, the diffusion lossThe method comprises the following steps: />The method comprises the steps of carrying out a first treatment on the surface of the Wherein->For calculating the length of the vector; by means of a loss function->Calculating the loss->:/>The method comprises the steps of carrying out a first treatment on the surface of the Loss functionIs a cross entropy loss function.
S5, dialogingThe specific process of inputting a dialogue abstract model and generating the dialogue abstract comprises the following steps:
s51: based on the input dialog, using a query generation moduleGenerating corresponding inquiry information->The method comprises the steps of carrying out a first treatment on the surface of the The scheme of the query generation module for generating the query information is prior art, in particular the query generation module comprises a query encoder and a query decoder, which will talk +.>An input query encoder outputting a query dialogue vector; for->The query dialogue vector and the query generation module are generated before->Personal word->Feeding into an interrogation encoder to obtain the +.>Personal word->And then all predicted words of the query generation module are obtained +.>Obtain inquiry information->The method comprises the steps of carrying out a first treatment on the surface of the When->When (I)>,/>Is the character marking the beginning of the abstract.
S52: will beAnd->Concatenation, input dialogue encoder, get query perceived dialogue characterization +.>;
S53: generating a noise set by a noise binding module, randomly sampling from the noise set to obtain a noise vectorWill->Initial state as dialog digest generation procedure +.>Use of a diffusion codec pair +.>Iterative denoising is carried out to obtain binary bit representation corresponding to the dialogue abstract>Will->Converting into decimal system to obtain final dialog abstract +.>。
In step S53, when generating the noise set through the noise binding module, the method specifically includes:
dialog for a conversationThe vocabulary of all word compositions in +.>The vocabulary of all word components in the manually annotated dialog abstract is +.>The words which do not appear in the dialogue abstract are all regarded as noise information and are treated as word list +.>He vocabulary->Collecting difference set operation to obtain universal binding noise set +.>;
Recording the manually marked user inquiry set asWherein->Representing the user query set +.>The M-th individual query of (1) remembers the word list of the word composition in all individual queries as +.>Words which do not appear in the individual queries are regarded as noise information by means of the vocabulary +.>He vocabulary->Collecting difference set operation to obtain interactive perception binding noise set;
The final noise set is obtained by carrying out union operation on the universal binding noise set and the interactive perception binding noise set。
In step S53, the finally generated dialogue abstract is obtainedThe specific process of (2) comprises:
random sampling from noise setsEach noise word is converted into a noise representation by using an embedding layer matrix, and then all the noise representations are averaged and normalized to obtain a noise vector +.>Will->Initial state as dialog digest generation procedure +.>;
Diffusion decoderFor->Performing iterative denoising, wherein the iterative process comprises->Step->The characteristic of the denoising ratio obtained after the step denoising is marked as +.>First->Characterization of the noise ratio obtained in the step->;
Wherein the intermediate variable
;
Representing a predefined hyper-parameter;
characterization of the resulting binary bitsConverting into decimal system to obtain final dialog abstract +.>。
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.
Furthermore, it should be understood that although the present disclosure describes embodiments, not every embodiment is provided with a single embodiment, and that this description is provided for clarity only, and that the disclosure is not limited to specific embodiments, and that the embodiments may be combined appropriately to form other embodiments that will be understood by those skilled in the art.
Claims (4)
1. Dialogue abstracting method based on noise binding diffusion modelInputting a dialogue abstract model to generate a dialogue abstract; the dialogue abstract model comprises a noise binding module, a query generation module, a dialogue encoder and a diffusion encoder-decoder; the diffusion encoder-decoder includes a diffusion encoder and a diffusion decoder;
a training process for a dialog summary model, comprising the steps of:
s1, using a query generation module, according to the input dialogueGenerating corresponding inquiry information->;
S2, inquiring the informationAnd dialogue->Concatenation, input dialogue encoder, get query perceived dialogue characterization +.>;
S3, randomly generating noise vectors through a noise binding moduleAnd abstracts the manually marked dialog +.>Bit representation converted into binary>Diffusion encoder uses noise vector +>For->Performing diffusion noise treatment, wherein the diffusion noise treatment process comprises +.>Step, obtaining the noise adding representation of the diffusion noise adding treatment step t->And noise characterization of the last step +.>;
S4, characterizing the dialogueInput diffusion encoder-decoder and use diffusion decoder +.>Characterization of noise->Performing denoising treatment comprising ∈>Step, get the denoising treatment +.>Step De-N ratio characterization->De-N ratio characterization of the last step->Predicted dialog abstract->Denoising treatment->Step De-N ratio characterization->Characterization of the noise obtained in step t ∈ ->Comparison gives diffusion loss->The method comprises the steps of carrying out a first treatment on the surface of the Predicted dialogue abstract->Dialog abstract with manual annotationIn contrast, calculate loss->The method comprises the steps of carrying out a first treatment on the surface of the Total loss->The method comprises the steps of carrying out a first treatment on the surface of the By total loss->And updating model parameters of the diffusion encoder-decoder using a back propagation algorithm; that is, only model parameters of the diffusion encoder-decoder are updated during dialogue digest model training;
in step S3, a noise vector is randomly generated through a noise binding moduleThe method specifically comprises the following steps: generating a noise set by means of a noise binding module, randomly sampling +.>A noise word, using an embedding layer matrix +.>Converting each noise word into a noise representation +.>All noise characterizations are then averaged and normalized +.>Obtaining noise vector->;/>Indicate->A noise characterization;
in step S3, the artificially marked dialogue abstract is summarizedBit representation converted into binary>Using noise vectors->For a pair ofPerforming diffusion noise adding treatment to obtain noise adding representation->The method specifically comprises the following steps: dialogue abstract based on bit-spreading network>Bit representation of the word in (a) converted into binary>Then use the noise vector +.>For->Performing diffusion noise treatment, wherein the diffusion noise treatment process comprises +.>Step (2) diffusion noise treatment the noise characterization obtained in step (t)>The method comprises the following steps:;/>representing a predefined hyper-parameter;
in step S4, the diffusion lossThe method comprises the following steps: />The method comprises the steps of carrying out a first treatment on the surface of the Wherein->For calculating the length of the vector; by means of a loss function->Calculating the loss->:/>The method comprises the steps of carrying out a first treatment on the surface of the Loss functionIs a cross entropy loss function.
2. The dialog summarization method based on the noise-binding diffusion model according to claim 1, wherein: will talk toThe specific process of inputting a dialogue abstract model and generating the dialogue abstract comprises the following steps:
s51: based on the input dialog, using a query generation moduleGenerating corresponding inquiry information->;
S52: will beAnd->Concatenation, input dialogue encoder, get query perceived dialogue characterization +.>;
S53: generating a noise set by a noise binding module, randomly sampling from the noise set to obtain a noise vectorWill->Initial state as dialog digest generation procedure +.>Use of a diffusion codec pair +.>Iterative denoising is carried out to obtain the dialogue abstract corresponding toBinary bit representation->Will->Converting into decimal system to obtain final dialog abstract +.>。
3. The method for dialogue summarization based on the noise binding diffusion model according to claim 2, wherein in step S53, when generating the noise set by the noise binding module, specifically comprising:
dialog for a conversationThe vocabulary of all word compositions in +.>The vocabulary of all word components in the manually marked dialogue abstract isThe words which do not appear in the dialogue abstract are all regarded as noise information and are treated as word list +.>He vocabulary->Collecting difference set operation to obtain universal binding noise set +.>;
Recording the manually marked user inquiry set asWherein, the method comprises the steps of, wherein,/>representing the user query set +.>The M-th individual query of (1) remembers the word list of the word composition in all individual queries as +.>Words which do not appear in the individual queries are regarded as noise information by means of the vocabulary +.>He vocabulary->Collecting difference set operation to obtain interactive perception binding noise set;
The final noise set is obtained by carrying out union operation on the universal binding noise set and the interactive perception binding noise set。
4. The method of dialogue summarization based on noise-binding diffusion model according to claim 2, wherein in step S53, a finally generated dialogue summary is obtainedThe specific process of (2) comprises:
random sampling from noise setsEach noise word is converted into a noise representation using an embedding layer matrix, and then all the noise representations are averaged and normalizedOperation, get noise vector->Will->Initial state as dialog digest generation procedure +.>;
Diffusion decoderFor->Performing iterative denoising, wherein the iterative process comprises->Step->The characteristic of the denoising ratio obtained after the step denoising is marked as +.>First->Characterization of the noise ratio obtained in the step->;
Wherein the intermediate variable
;
Representing a predefined hyper-parameter;
characterization of the resulting binary bitsConverting into decimal system to obtain final dialog abstract +.>。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311395915.2A CN117131187B (en) | 2023-10-26 | 2023-10-26 | Dialogue abstracting method based on noise binding diffusion model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311395915.2A CN117131187B (en) | 2023-10-26 | 2023-10-26 | Dialogue abstracting method based on noise binding diffusion model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117131187A CN117131187A (en) | 2023-11-28 |
CN117131187B true CN117131187B (en) | 2024-02-09 |
Family
ID=88858623
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311395915.2A Active CN117131187B (en) | 2023-10-26 | 2023-10-26 | Dialogue abstracting method based on noise binding diffusion model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117131187B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2008310784A (en) * | 2007-06-13 | 2008-12-25 | Shinichiro Okude | Interaction device using semantic network as base, interaction program, and recording medium recording interaction program |
US10152970B1 (en) * | 2018-02-08 | 2018-12-11 | Capital One Services, Llc | Adversarial learning and generation of dialogue responses |
CN115017299A (en) * | 2022-04-15 | 2022-09-06 | 天津大学 | Unsupervised social media summarization method based on de-noised image self-encoder |
CN115794999A (en) * | 2023-02-01 | 2023-03-14 | 北京知呱呱科技服务有限公司 | Patent document query method based on diffusion model and computer equipment |
CN115905513A (en) * | 2023-02-22 | 2023-04-04 | 中国科学技术大学 | Dialog summarization method based on denoising type question answering |
CN116484858A (en) * | 2023-04-19 | 2023-07-25 | 南京邮电大学 | Text abstract generation method based on diffusion model |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110188182B (en) * | 2019-05-31 | 2023-10-27 | 中国科学院深圳先进技术研究院 | Model training method, dialogue generating method, device, equipment and medium |
US20230109379A1 (en) * | 2021-10-05 | 2023-04-06 | Nvidia Corporation | Diffusion-based generative modeling for synthetic data generation systems and applications |
-
2023
- 2023-10-26 CN CN202311395915.2A patent/CN117131187B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2008310784A (en) * | 2007-06-13 | 2008-12-25 | Shinichiro Okude | Interaction device using semantic network as base, interaction program, and recording medium recording interaction program |
US10152970B1 (en) * | 2018-02-08 | 2018-12-11 | Capital One Services, Llc | Adversarial learning and generation of dialogue responses |
CN115017299A (en) * | 2022-04-15 | 2022-09-06 | 天津大学 | Unsupervised social media summarization method based on de-noised image self-encoder |
CN115794999A (en) * | 2023-02-01 | 2023-03-14 | 北京知呱呱科技服务有限公司 | Patent document query method based on diffusion model and computer equipment |
CN115905513A (en) * | 2023-02-22 | 2023-04-04 | 中国科学技术大学 | Dialog summarization method based on denoising type question answering |
CN116484858A (en) * | 2023-04-19 | 2023-07-25 | 南京邮电大学 | Text abstract generation method based on diffusion model |
Also Published As
Publication number | Publication date |
---|---|
CN117131187A (en) | 2023-11-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2019174450A1 (en) | Dialogue generation method and apparatus | |
CN110619043A (en) | Automatic text abstract generation method based on dynamic word vector | |
CN111666756B (en) | Sequence model text abstract generation method based on theme fusion | |
CN116072098B (en) | Audio signal generation method, model training method, device, equipment and medium | |
CN107832300A (en) | Towards minimally invasive medical field text snippet generation method and device | |
CN114547293A (en) | Cross-platform false news detection method and system | |
CN115905513B (en) | Dialogue abstracting method based on denoising type question and answer | |
Wang et al. | Residual-guided personalized speech synthesis based on face image | |
CN117131187B (en) | Dialogue abstracting method based on noise binding diffusion model | |
CN114942990A (en) | Few-sample abstract dialogue abstract generation system based on prompt learning | |
Xu et al. | Poke: Prior knowledge enhanced emotional support conversation with latent variable | |
CN115525743B (en) | Man-machine interaction method based on self-learning whitening network and electronic equipment | |
Zhang et al. | Voice conversion with denoising diffusion probabilistic gan models | |
CN116467452A (en) | Chinese complaint classification method based on multi-task learning hybrid neural network | |
Jiang et al. | Pednet: A persona enhanced dual alternating learning network for conversational response generation | |
CN110347824B (en) | Method for determining optimal number of topics of LDA topic model based on vocabulary similarity | |
Mustafa et al. | Framewise WaveGAN: High speed adversarial vocoder in time domain with very low computational complexity | |
Ning et al. | Dualvc 2: Dynamic masked convolution for unified streaming and non-streaming voice conversion | |
CN116561259B (en) | Method, device, equipment and medium for testing session segmentation model | |
JP2020053015A (en) | Interest estimation method, interest estimation device and interest estimation program | |
US20230283496A1 (en) | Engagement analysis for remote communication sessions | |
CN117174240B (en) | Medical image report generation method based on large model field migration | |
CN108959479B (en) | Event emotion classification method based on text similarity | |
Chen et al. | A noise-robust voice conversion method with controllable background sounds | |
KR20230146398A (en) | Sequence text summary processing device using bart model and control method thereof |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |