CN111597811A - Financial chapter-level multi-correlation event extraction method based on graph neural network algorithm - Google Patents
Financial chapter-level multi-correlation event extraction method based on graph neural network algorithm Download PDFInfo
- Publication number
- CN111597811A CN111597811A CN202010394858.6A CN202010394858A CN111597811A CN 111597811 A CN111597811 A CN 111597811A CN 202010394858 A CN202010394858 A CN 202010394858A CN 111597811 A CN111597811 A CN 111597811A
- Authority
- CN
- China
- Prior art keywords
- event
- sentence
- elements
- gcnn
- neural network
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
- G06F40/295—Named entity recognition
-
- 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
- G06F16/35—Clustering; Classification
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Databases & Information Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a financial chapter-level multi-association event extraction method based on a graph neural network algorithm, which comprises the steps of firstly segmenting input sentences, and extracting event candidate elements through bert + crf; secondly, constructing a GCNN node, and splicing the representation of the event candidate element, the element position code, the sentence representation of the event candidate element and the sentence position code into a node representation; then, constructing a GCNN edge; secondly, updating the representation of the event candidate elements based on a GCNN network, and obtaining event types and event role classification results corresponding to the elements through a linear layer and a multi-sigmoid layer; and finally, outputting the extracted information. The graph convolution neural network applied by the invention makes up the information of the correlation of the multiple event elements between sentences which can not be captured in the existing other chapter-level event extraction methods, and the adoption of the algorithm structure of bert + CRF can more accurately and more comprehensively describe the node vector in GCNN and the corresponding sentence vector.
Description
Technical Field
The invention relates to the technical field of information, in particular to a financial chapter-level multi-correlation event extraction method based on a graph neural network algorithm.
Background
The main research methods for event extraction include pattern matching and machine learning. The pattern matching can achieve higher performance in a specific field, but the portability is poor. In the extraction method of machine learning, the machine learning is irrelevant to the field, the guidance of too many field experts is not needed, and the system portability is good.
1. Event extraction based on pattern matching
The pattern matching method is to perform the identification and extraction of events under the guidance of some patterns. The mode is mainly used for indicating a context constraint ring forming target information, and the fusion of domain knowledge and language knowledge is intensively embodied. During extraction, information meeting the pattern constraint condition is found out through various pattern matching algorithms. Therefore, the core of the method is the construction of the extraction mode. Typical pattern matching-based event extraction systems are ExDisco, GenPAM, and the like. Initially, the patterns were established primarily by manual methods. According to the method, the automatic acquisition of the patterns in the doctor thesis of Jiangji of Chinese academy of sciences is deeply researched, and the GenPAM (GenPAM) event extraction pattern learning method based on the domain-independent concept knowledge base is provided.
Generally, the method based on pattern matching can achieve better effect in a specific field, but the portability of the system is poor, and when the method is transplanted from one field to another field, the pattern needs to be reconstructed. And the construction of the mode is time-consuming and labor-consuming, and requires the guidance of domain experts. Although the introduction of the machine learning method can speed up the acquisition of the patterns to some extent, the resulting conflict between different patterns is also a troublesome problem. Moreover, most semantic degrees of the existing research are on a syntax level, and the semantic degree needs to be further improved.
2. Event extraction based on machine learning
The event is identified by adopting a machine learning method, namely the idea of text classification is used for reference, the identification of event types and event elements is converted into a classification problem, and the core of the method lies in the structure of a classifier and the selection of characteristics. However, the event classification is different from the text classification, and mainly appears in the following aspects: the classified texts are short, and most of the classified texts are complete sentences; since the sentence is an event expression sentence, the amount of information included in the sentence is large.
Most of researches are based on trigger words to detect events, the method is simple and intuitive, but the trigger words only account for a small part of all words, so that a large number of counter-examples are introduced in training, the counter-examples are unbalanced, and the judgment of each word causes additional increase of calculation amount. In order to solve the above problem, a method based on a combination of trigger expansion and binary classification has been studied to identify the event category. The trigger words are recorded in the dictionary during training and are expanded through the synonym forest, so that the problems of unbalanced positive and negative examples and data sparseness of training examples are solved well, and a good effect is achieved on Chinese linguistic data of the ACE.
In summary, the method based on machine learning does not depend on the content and format of the corpus, but needs a large-scale standard corpus, otherwise, a serious data sparsity problem occurs. However, the corpus scale at the present stage is difficult to meet the application requirement, and manual corpus tagging is time-consuming and labor-consuming, and in order to alleviate the difficulty in acquiring tagged corpora, relevant learners have explored semi-supervised and unsupervised learning research. In addition, feature selection is also an important factor for determining the quality of the machine learning result. Therefore, how to avoid the data sparseness phenomenon and how to select an appropriate feature become important issues for machine learning method based research. Most of current research is based on phrase or sentence level information, and it will become a new hotspot to improve extraction performance by using chapter level or cross-chapter information.
In the existing financial chapter-level multi-association event extraction method, the following three problems mainly exist:
there are many issues: there are a number of complex extraction tasks in financial bulletin data where multiple events are interrelated, which may involve the same entity but not the same role as it is in the respective event, or there is no shared entity between the events but there is a causal or chronological association of events. These conditions are the factors that cause the extraction of multiple events to be very difficult at present, and the prior art can not deal with the common problems with the great difficulty.
Information dispersion problem: in the chapter-level multi-event extraction task, different events are often dispersed in different sentences and are far apart. In this case, it is difficult for the prior art to take the information related to the dispersion into consideration, and the information on one side affects the accuracy of the multiple event extraction.
Event types and argument information are generally extracted in series in the prior art, which can cause information loss, accuracy reduction and longer processing time.
Disclosure of Invention
The invention aims to overcome the technical problems and provides a financial chapter-level multi-correlation event extraction method based on a graph neural network algorithm.
In order to achieve the purpose, the invention adopts the following technical scheme:
a financial chapter-level multi-correlation event extraction method based on a graph neural network algorithm comprises the following steps:
s1: segmenting the input sentence, and extracting event candidate elements through bert + crf;
s2: constructing a GCNN node: obtaining the representation of the event candidate element, the element position code, the sentence representation of the event candidate element and the sentence position code, and splicing the representation into a node representation;
s3: constructing a GCNN edge;
s4: updating the characterization of the event candidate elements based on a GCNN, and obtaining event types and event role classification results corresponding to the elements through a linear layer and a multi-sigmoid layer;
s5: and outputting the extracted information.
Further, the specific method for constructing the GCNN edge in step S3 is as follows:
(1) and judging whether the sentence vector contains event information after passing through a linear layer and a sigmoid activation layer, and if so, taking the sentence vector as a candidate sentence for constructing a graph convolution network edge:
(2) shared entity names exist in the connection candidate sentences, wherein the shared entity names comprise common nouns or pronouns of business names and person names;
(3) for entity names that have a share, it is bordered by other event candidate elements that appear within the same sentence.
Further, the step S4 is to classify the event type and the event role through a joint model.
Further, the specific steps of classifying the event type and the event role through the combined model are as follows:
(1) dividing the updated event candidate element set X into two types, wherein one type is event key information, and the other type is event supplementary information;
(2) for the elements of the event supplementary information class, the event roles of the elements can be classified only by judging the event types of the elements; and for the elements of the event key information class, a joint model is required to realize classification.
Further, in the combined model, the structures of the event type classifier Ct and the event role classifier Cr are composed of a linear layer and p sigmoid layers, where p is 4 in the event type classifier and 19 in the event role classifier, and the classification result of the event type in the combined model affects the classification result of the event role.
The invention has the beneficial effects that:
1. the graph convolution neural network applied by the invention makes up the information of the correlation of the multiple event elements between sentences which can not be captured in the existing other chapter-level event extraction methods. The graph convolutional neural network can learn the spatial topology structure between network nodes in the non-European space, and the graph convolutional neural network has obvious advantages for the task of information extraction which is not limited to the upper sentence and the lower sentence. In addition, with the accumulation of the number of layers of the GCNN, the node association information at a deeper layer in the network can be learned, which is a place that cannot be realized by other existing methods. Therefore, the GCNN-based algorithm has obvious advantages aiming at the difficulties of the complicated association and the information dispersion of events in the financial bulletin multi-event extraction task.
2. In order to accurately extract event element information, an algorithm structure of bert + CRF is adopted; and in consideration of the huge pre-training corpus and the intra-sentence context semantic representation capability of bert, the node vectors and the corresponding sentence vectors in the GCNN can be more accurately and more comprehensively described.
3. In the event type (trigger word) classification and event role classification tasks, considering that the category of an event role depends on the type of an event, the classification result of the event role is also possible to reversely change the classification of the event type, the overall accuracy of the event type and the event role is improved by adopting a combined model, and the time cost in prediction is also reduced.
Drawings
FIG. 1: the work flow of the invention is shown schematically.
FIG. 2: the invention relates to a schematic diagram of event correlation vector transformation.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and examples.
Example one
The financial bulletins entered are: on a certain month and a certain month of the year, the YY company, the Shandong province company and the ZX bank company Jinan division sign enter into a 'entrusted loan contract', namely, the ZX bank company entrusts the ZX bank Jinan division to provide loan of not more than 3 hundred million yuan for the YY company, and the loan period is 24 months. Secondly, because the loan term is expired, the two parties plan to sign a debt reorganization agreement through negotiation between YY stock limited company and a certain stock limited company in Shandong province, the two parties confirm that the 3 hundred million yuan loan principal is the reorganized debt principal under the agreement, and the debt reorganization grace period is 1 year. The Yoy stock company is subordinate to Ningbo YY real estate development company, Renbell YY real estate development company and Ningbo RY industry establishment company, and is intended to provide guarantee for the recombination of the debts, the Ningbo YY real estate development company is intended to provide quality guarantee for the recombination of the debts by 100% of the stock right of the Renbell YY real estate development company, the Renbell YY real estate development company is intended to provide mortgage guarantee for the recombination of the debts by using part of the assets of the Renbell YY real estate development company, the guarantee amount is 3 hundred million yuan, and the guarantee period is two years from the day when the recombination of the debts is expired.
As shown in fig. 1, first, the financial bulletin is divided into N sentences in sentence units, and the upper limit of the number of words in each sentence is M;
then, all sentences are passed through a model of Bert + CRF to extract event candidate elements including business name (COPR), person name (PER), amount (MON), Share (SHA) and Term (TIM), respectively, as shown in fig. 2b as E1-E4. Meanwhile, Bert can encode a sentence representation vector, wherein the number of hidden layers is H, such as S1-S4;
then, after the vector of sentence S1 obtained in fig. 2b is subjected to the maximum pooling operation, the vector space is changed from M × H to H; similarly, the vector space becomes H after the word vectors included in E1 are subjected to average pooling. In addition, the position of each sentence in the chapter and the position of the event candidate element appearing in each sentence are also important features affecting the extraction of the event information, so as shown in fig. 2c, the pooled S1 vector and its position-coding information, and the pooled E1 and its position-coding information are respectively spliced together to serve as the node input information of the graph convolution network.
Then, constructing a GCNN edge, wherein the specific construction steps are as follows:
(1) judging whether the sentence contains event information after the sentence vector passes through a linear layer and a sigmoid activation layer, and if so, taking the sentence as a candidate sentence for constructing a graph convolution network edge;
(2) shared business names and person names exist in the connection candidate sentences, and the shared business names and the person names comprise common nouns or pronouns of the business names and the person names.
(3) For business name and person name entities that have a share, it is bordered by other event candidate elements that appear within the same sentence.
Then, according to the nodes of the constructed graph convolution network and the rules of the GCNN edges, an undirected graph is defined:
G＝(ν，)
and the iterative update is realized according to the following formula
Wherein the content of the first and second substances,
vector representation representing the ith event candidate element in the kth module of the GCNN, v (i) representing all the neighboring nodes of node i,
and
are all parameters of the edge between the node i and the node u in the k-th module of the GCNN.
The association information between nodes which are far away is accumulated by overlapping K GCNN modules, and the information transfer between adjacent nodes is controlled by using an edge-based gating mechanism.
Then, after the processing by the graph convolution network, the characterizations of all event candidate elements are updated, and the updated representations are represented as E1 'to E4' in FIG. 2E, and the set is represented as χ.
Then, in order to improve the efficiency of extracting the event information, the event type and the event role are classified through a combined model, and the specific steps are as follows:
(1) and dividing the updated event candidate element set chi into two types, wherein one type is event key information, namely enterprise name and person name, and the other type is event supplementary information, namely money, shares and period.
(2) For the elements of the event supplementary information class, the event roles which the elements serve can be classified only after judging the event types of the elements. And for the elements of the event key information class, a joint model is required to realize classification.
The structure of the event type classifier Ct and the event role classifier Cr is composed of a linear layer and p sigmoid layers, wherein p is 4 in the event type classifier, and p is 19 in the event role classifier. The classification result of the event type in the combined model can influence the classification result of the event role, and the logic is also met in practical application.
The combined model of the event type and the event role classification satisfies the following formula when calculating the loss value:
losstotal＝lossCt+lossCr
wherein lossCtRepresents the overall loss, of the event type classifierCrRepresents the overall loss of the event role classifier, and losstotalIs the loss value we will eventually optimize.
And finally, outputting the extraction result.
The following briefly introduces the extraction process of this embodiment:
the announcement relates to 4 types of events, the 4 types of events are related and associated, and information is dispersed, and the extraction process according to the method comprises the following steps:
1. the bulletin is broken into 3 sentences, and event elements such as business names are identified by bert + crf: YY corporation, ZX bank corporation, jonan division, ningbo YY real estate development corporation, trenebel YY real estate development corporation, ningbo RY placement corporation, in amounts: 3 billion yuan, term: 24 months, etc.
2. Constructing a GCNN node: and obtaining the representation of the event element, the element position code, the representation of the sentence where the event element is located and the sentence position code, and then splicing the representation of the event element, the element position code, the representation of the sentence where the event element is located and the sentence position code into a node representation.
3. Constructing a GCNN edge: the bank can be used for storing YY-ZX bank, YY-YY, YY-Ningbo YY, YY-Relenbel YY, YY-RY and the like.
4. And updating the representation of the event elements based on the GCNN, and obtaining the event type and the event role classification result corresponding to each element through a linear layer and a multi-sigmoid layer.
5. The finally extracted event information is shown in table 1.
TABLE 1
Finally, it should be noted that: the above embodiments are only used to illustrate the present invention and do not limit the technical solutions described in the present invention; thus, while the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted; all such modifications and variations are intended to be included herein within the scope of this disclosure and the present invention and protected by the following claims.
Claims (5)
1. A financial chapter-level multi-correlation event extraction method based on a graph neural network algorithm is characterized by comprising the following steps:
s1: segmenting the input sentence, and extracting event candidate elements through bert + crf;
s2: constructing a GCNN node: obtaining the representation of the event candidate element, the element position code, the sentence representation of the event candidate element and the sentence position code, and splicing the representation into a node representation;
s3: constructing a GCNN edge;
s4: updating the characterization of the event candidate elements based on a GCNN, and obtaining event types and event role classification results corresponding to the elements through a linear layer and a multi-sigmoid layer;
s5: and outputting the extracted information.
2. The method for extracting financial chapter-level multi-correlation events based on graph neural network algorithm of claim 1, wherein the specific method for constructing the GCNN side in step S3 is as follows:
(1) judging whether the sentence contains event information after the sentence vector passes through a linear layer and a sigmoid activation layer, and if so, taking the sentence as a candidate sentence for constructing a graph convolution network edge;
(2) shared entity names exist in the connection candidate sentences, wherein the shared entity names comprise common nouns or pronouns of business names and person names;
(3) for entity names that have a share, it is bordered by other event candidate elements that appear within the same sentence.
3. The method for extracting financial chapter-level multi-correlation events based on graph neural network algorithm of claim 1, wherein: the step S4 is to classify the event type and the event role through a joint model.
4. The method for extracting financial chapter-level multi-correlation events based on graph neural network algorithm of claim 3, wherein the specific steps of classifying the event type and the event role through the combined model are as follows:
(1) dividing the updated event candidate element set X into two types, wherein one type is event key information, and the other type is event supplementary information;
(2) for the elements of the event supplementary information class, the event roles of the elements can be classified only by judging the event types of the elements; and for the elements of the event key information class, a joint model is required to realize classification.
5. The method for extracting financial chapter-level multi-correlation events based on graph neural network algorithm of claim 4, wherein: in the combined model, the structures of the event type classifier Ct and the event role classifier Cr are composed of a linear layer and p sigmoid layers, p is 4 in the event type classifier, p is 19 in the event role classifier, and the classification result of the event type in the combined model influences the classification result of the event role.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010394858.6A CN111597811B (en) | 2020-05-09 | 2020-05-09 | Financial chapter-level multi-correlation event extraction method based on graph neural network algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010394858.6A CN111597811B (en) | 2020-05-09 | 2020-05-09 | Financial chapter-level multi-correlation event extraction method based on graph neural network algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111597811A true CN111597811A (en) | 2020-08-28 |
| CN111597811B CN111597811B (en) | 2021-11-12 |
Family
ID=72191135
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010394858.6A Active CN111597811B (en) | 2020-05-09 | 2020-05-09 | Financial chapter-level multi-correlation event extraction method based on graph neural network algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111597811B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112328801A (en) * | 2020-09-28 | 2021-02-05 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Method for predicting group events by event knowledge graph |
| CN113282726A (en) * | 2021-05-27 | 2021-08-20 | 成都数之联科技有限公司 | Data processing method, system, device, medium and data analysis method |
| CN116245139A (en) * | 2023-04-23 | 2023-06-09 | 中国人民解放军国防科技大学 | Training method and device for graph neural network model, event detection method and device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951438A (en) * | 2017-02-13 | 2017-07-14 | 北京航空航天大学 | A kind of event extraction system and method towards open field |
| CN109325228A (en) * | 2018-09-19 | 2019-02-12 | 苏州大学 | English event trigger word abstracting method and system |
| CN110134720A (en) * | 2019-05-17 | 2019-08-16 | 苏州大学 | It merges local feature and combines abstracting method with the event of deep learning |
| CN110377738A (en) * | 2019-07-15 | 2019-10-25 | 昆明理工大学 | Merge the Vietnamese news event detecting method of interdependent syntactic information and convolutional neural networks |
| CN110765265A (en) * | 2019-09-06 | 2020-02-07 | 平安科技（深圳）有限公司 | Information classification extraction method and device, computer equipment and storage medium |
-
2020
- 2020-05-09 CN CN202010394858.6A patent/CN111597811B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951438A (en) * | 2017-02-13 | 2017-07-14 | 北京航空航天大学 | A kind of event extraction system and method towards open field |
| CN109325228A (en) * | 2018-09-19 | 2019-02-12 | 苏州大学 | English event trigger word abstracting method and system |
| CN110134720A (en) * | 2019-05-17 | 2019-08-16 | 苏州大学 | It merges local feature and combines abstracting method with the event of deep learning |
| CN110377738A (en) * | 2019-07-15 | 2019-10-25 | 昆明理工大学 | Merge the Vietnamese news event detecting method of interdependent syntactic information and convolutional neural networks |
| CN110765265A (en) * | 2019-09-06 | 2020-02-07 | 平安科技（深圳）有限公司 | Information classification extraction method and device, computer equipment and storage medium |
Non-Patent Citations (4)
| Title |
|---|
| THIEN HUU NGUYEN等: "Joint Event Extraction via Recurrent Neural Networks", 《PROCEEDINGS OF NAACL-HLT 2016》 * |
| XIAO LIU等: "Jointly Multiple Events Extraction via Attention-based Graph Information Aggregation", 《HTTPS://ARXIV.ORG/ABS/1809.09078》 * |
| YUBO CHEN等: "Event Extraction via Dynamic Multi-Pooling Convolutional Neural Networks", 《PROCEEDINGS OF THE 53RD ANNUAL MEETING OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICSAND THE 7TH INTERNATIONAL JOINT CONFERENCE ON NATURAL LANGUAGE PROCESSING》 * |
| 仲伟峰 等: "基于联合标注和全局推理的篇章级事件抽取", 《中文信息学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112328801A (en) * | 2020-09-28 | 2021-02-05 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Method for predicting group events by event knowledge graph |
| CN112328801B (en) * | 2020-09-28 | 2022-06-14 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Method for predicting group events by event knowledge graph |
| CN113282726A (en) * | 2021-05-27 | 2021-08-20 | 成都数之联科技有限公司 | Data processing method, system, device, medium and data analysis method |
| CN116245139A (en) * | 2023-04-23 | 2023-06-09 | 中国人民解放军国防科技大学 | Training method and device for graph neural network model, event detection method and device |
| CN116245139B (en) * | 2023-04-23 | 2023-07-07 | 中国人民解放军国防科技大学 | Training method and device for graph neural network model, event detection method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111597811B (en) | 2021-11-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109783818B (en) | Enterprise industry classification method | |
| CN110020438B (en) | Sequence identification based enterprise or organization Chinese name entity disambiguation method and device | |
| CN111597811B (en) | Financial chapter-level multi-correlation event extraction method based on graph neural network algorithm | |
| CN112417877B (en) | Text inclusion relation recognition method based on improved BERT | |
| US20230075341A1 (en) | Semantic map generation employing lattice path decoding | |
| CN111325029B (en) | Text similarity calculation method based on deep learning integrated model | |
| Cui et al. | Sentiment analysis via integrating distributed representations of variable-length word sequence | |
| CN112434535A (en) | Multi-model-based factor extraction method, device, equipment and storage medium | |
| CN114491024B (en) | Specific field multi-label text classification method based on small sample | |
| CN111709225B (en) | Event causal relationship discriminating method, device and computer readable storage medium | |
| CN114153978A (en) | Model training method, information extraction method, device, equipment and storage medium | |
| CN111191051A (en) | Method and system for constructing emergency knowledge map based on Chinese word segmentation technology | |
| CN111209362A (en) | Address data analysis method based on deep learning | |
| CN114840685A (en) | Emergency plan knowledge graph construction method | |
| CN109543038A (en) | A kind of sentiment analysis method applied to text data | |
| Zhang et al. | A text mining based method for policy recommendation | |
| Yang et al. | A general solution and practice for automatically constructing domain knowledge graph | |
| Borisov et al. | Method for documents rubrication and analysis based on fuzzy relations of difference between their syntactical characteristics | |
| CN112948561A (en) | Method and device for automatically expanding question-answer knowledge base | |
| Wang et al. | Preprocessing and feature extraction methods for microfinance overdue data | |
| Ramos Flores et al. | Proposal for named entities recognition and classification (NERC) and the automatic generation of rules on Mexican news | |
| CN111046934A (en) | Method and device for identifying soft clauses of SWIFT message | |
| CN109299442A (en) | Chinese chapter primary-slave relation recognition methods and system | |
| CN117291192B (en) | Government affair text semantic understanding analysis method and system | |
| He et al. | An Association Rule Mining Method Based on Named Entity Recognition and Text Classification |
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 |