CN112651243B - Abbreviated project name identification method based on integrated structured entity information and electronic device - Google Patents
Abbreviated project name identification method based on integrated structured entity information and electronic device Download PDFInfo
- Publication number
- CN112651243B CN112651243B CN202011481330.9A CN202011481330A CN112651243B CN 112651243 B CN112651243 B CN 112651243B CN 202011481330 A CN202011481330 A CN 202011481330A CN 112651243 B CN112651243 B CN 112651243B
- Authority
- CN
- China
- Prior art keywords
- encoder
- belonging
- classification probability
- entity
- abbreviated
- 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
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000012360 testing method Methods 0.000 claims abstract description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 abstract description 11
- 238000013135 deep learning Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000002372 labelling Methods 0.000 description 3
- 238000003058 natural language processing Methods 0.000 description 2
- 238000004806 packaging method and process Methods 0.000 description 2
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000009193 crawling Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
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
- G06F16/355—Class or cluster creation or modification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/205—Parsing
- G06F40/216—Parsing using statistical methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/237—Lexical tools
- G06F40/242—Dictionaries
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Probability & Statistics with Applications (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an abbreviated project name identification method based on merging structured entity information and an electronic device, comprising the following steps: acquiring a knowledge base anchor text, a common project name dictionary and abbreviated project names, and training a pre-training encoder by combining an entity boundary recognition module, a named entity extraction module and an abbreviated project name recognition module to obtain an abbreviated project name recognition model; inputting a test text into the abbreviated item name recognition model, and recognizing abbreviated item names in the test text. The invention improves the recognition rate and recall rate of the abbreviated item names in the text.
Description
Technical Field
The invention belongs to the field of natural language processing, and particularly relates to an abbreviated project name identification method based on merging structured entity information and an electronic device.
Background
The process of understanding natural language by humans can be seen as a process of extracting information from natural language text, followed by a summary of the information. The extraction information corresponds to an information extraction task in natural language processing, namely, a text processing technology for extracting the fact information of entities, relations, events and the like of a specified type from natural language texts and forming structured data output. The named entity identifies sub-problems belonging to the information extraction field, aims at identifying the named entity from unstructured input text, and generally comprises identifying the boundary of the entity and determining the entity type. And abbreviated project name recognition is a task of named entity recognition that limits the types of entities to project names and the existence of abbreviations in the sequence of entities.
The existing named entity recognition method is mainly based on a deep learning method, and the framework of the named entity recognition method comprises an input distribution representation layer, a context coding layer and a decoding layer. Because named entity identification data needs professional labeling, the labeling workload is large, and the labeling data is not easy to obtain, researchers propose a series of deep learning methods to integrate external knowledge into the existing named entity identification method, so that the model performance is improved to the greatest extent on limited training data. For example, a large amount of unlabeled text is incorporated into the BERT: pre-training of Deep Bidirectional Transformers for Language Understanding (simply referred to as BERT), which aims to make the model better understand semantic information, structural information, etc. existing in natural language. Category description information of entity tags is incorporated into A Unified MRC Framework for Named Entity Recognition (abbreviated as MRC-NER) to enable a model to better understand semantic information of a current entity category.
However, in the abbreviated item name recognition scenario, because of the abbreviation phenomenon in the entity sequence, the syntax information and the semantic information which can be provided by the entity sequence are imperfect, and a greater challenge is brought to the boundary recognition of the abbreviated item name. The existing knowledge-incorporated deep learning methods such as BERT, MRC-NER, etc. cannot enhance the boundary recognition information by using the existing knowledge, and thus cannot be directly used for abbreviated item name recognition scenes.
Disclosure of Invention
In order to overcome the defect of the prior knowledge-incorporated deep learning named entity recognition method in a boundary recognition subtask under an abbreviated item name recognition scene, the invention provides an abbreviated item name recognition method and device based on the information of an incorporated structured entity, and the recognition rate and recall rate of the abbreviated item name are improved through entity boundary recognition, named entity extraction and abbreviated item name recognition.
The technical scheme adopted by the invention for solving the technical problems mainly comprises the following steps:
an abbreviated project name recognition method based on merging structured entity information comprises the following steps:
1) Marking entity sequences in the encyclopedia knowledge base to obtain knowledge base anchor texts, and constructing a common item name dictionary through collected common item names;
2) Inputting the knowledge base anchor text into a pre-training encoder, classifying the character strings output by the pre-training encoder according to the fact that the character string classification probability belonging to the entity sequence is as high as possible and the character string classification probability not belonging to the entity sequence is as low as possible, so as to train the pre-training encoder, and obtaining a first encoder;
3) Matching the text of the knowledge base anchor text with the common project names in the common project name dictionary, inputting the matching result into a first encoder, classifying the character strings output by the first encoder according to the character string classification probability belonging to the project name entity sequence as high as possible and the character string classification probability not belonging to the project name entity sequence as low as possible so as to train the first encoder and obtain a second encoder;
4) Inputting marked abbreviation item name recognition data into a second encoder, classifying the character strings output by the second encoder according to the fact that the character string classification probability belonging to the abbreviation item name entity sequence is as high as possible and the character string classification probability not belonging to the abbreviation item name entity sequence is as low as possible, so as to train the second encoder, and obtaining an abbreviation item name recognition model;
5) Inputting a test text into the abbreviated item name recognition model, and recognizing abbreviated item names in the test text.
Further, the encyclopedia knowledge base includes: wikipedia knowledge base.
Further, the entity sequence is obtained by:
1) Constructing a regular expression;
2) And matching texts of the encyclopedia knowledge base through the regular expression to obtain an entity sequence.
Further, the format of the regular expression includes: r' \ [ {2} (.
Further, the pre-training encoder is a pre-training BERT model.
Further, by using the softmax classifier, the strings output by the pre-training encoder, the strings output by the first encoder, and the strings output by the second encoder are classified.
Further, the softmax classifier includes: a softmax start position classifier and a softmax end position classifier.
Further, the softmax start position classifier and the softmax end position classifier classify the strings output by the pre-training encoder by the following strategy:
1) Classifying the character strings output by the pre-training encoder according to the fact that the classification probability of the beginning characters belonging to the entity sequences is as high as possible and the classification probability of the beginning character strings not belonging to the entity sequences is as low as possible;
2) And classifying the character strings output by the pre-training encoder by a softmax ending position classifier according to the fact that the classification probability of the ending character belonging to the entity sequence is as high as possible and the classification probability of the ending character string not belonging to the entity sequence is as low as possible.
Further, the softmax start position classifier and the softmax end position classifier classify the character string output by the first encoder by the following strategy:
1) Classifying the character strings output by the first encoder by a softmax starting position classifier according to the fact that the classification probability of the starting character belonging to the entity sequence of the project name is as high as possible and the classification probability of the starting character string not belonging to the entity sequence of the project name is as low as possible;
2) And classifying the character strings output by the first encoder by a softmax ending position classifier according to the fact that the classification probability of ending characters belonging to the entity sequence of the project name is as high as possible and the classification probability of ending character strings not belonging to the entity sequence of the project name is as low as possible.
Further, the softmax start position classifier and the softmax end position classifier classify the character string output by the second encoder by the following strategy:
1) Classifying the character strings output by the second encoder according to the fact that the classification probability of the initial character belonging to the abbreviated item name entity sequence is as high as possible and the classification probability of the initial character string not belonging to the abbreviated item name entity sequence is as low as possible;
2) And classifying the character strings output by the second encoder according to the fact that the classification probability of the ending character belonging to the entity sequence of the abbreviated item name is as high as possible and the classification probability of the ending character string not belonging to the entity sequence of the abbreviated item name is as low as possible.
A storage medium having a computer program stored therein, wherein the computer program is arranged to perform the method described above when run.
An electronic device comprising a memory having a computer program stored therein and a processor arranged to run the computer to perform the method described above.
Compared with the prior art, the method improves the recognition rate and recall rate of the abbreviated item names in the text through entity boundary recognition, named entity extraction and abbreviated item name recognition.
Drawings
FIG. 1 is a flow chart of the abbreviated project name recognition model training of the present invention.
Fig. 2 is a schematic diagram of a structured information collection module.
Fig. 3 is a schematic diagram of an entity boundary identification module.
FIG. 4 is a diagram of a named entity extraction module
FIG. 5 is a schematic diagram of an abbreviated project name recognition module
Detailed Description
The present invention will be further described with reference to the following examples, in order to make the above objects, features and advantages of the present invention more comprehensible.
As shown in fig. 1, the training process of the abbreviated project name recognition model of the present invention is composed of four modules:
and a structured information collection module. The method comprises the steps of collecting knowledge base anchor texts, commonly used project names and forming a commonly used project name dictionary.
And an entity boundary identification module. And according to the anchor text of the knowledge base, providing an entity boundary recognition algorithm to carry out algorithm preheating.
And naming entity extraction module. And according to the common project name dictionary, providing a named entity extraction algorithm for algorithm pre-training.
The abbreviation item name recognition module. And carrying out algorithm training according to the abbreviated item name recognition data marked by the expert, and packaging the algorithm into an abbreviated item name recognition model.
First part, structured information collection module (FIG. 2)
The structured information collection module of the invention mainly comprises the following two steps:
a) Knowledge base anchor text is collected. Downloading and decompressing the packaged wikipedia knowledge base, and marking the anchor text as an entity sequence. Wherein "anchor text" is a string shaped as "[ [ arbitrary string ] ]", the present invention designs a regular expression "r' \ [ {2} (.?.
b) The common item names are collected. And (3) crawling the webpage with the designed project name, and manually screening and denoising to obtain a common project name dictionary.
Second part, entity boundary identification module (FIG. 3)
Knowledge base anchor text labeled with the sequence of entities is used as input to the module, while the input "identify entities in the text" is given. And splicing the two sections of input and inputting the two sections of input into the entity boundary recognition module. The main algorithms of the module are a pre-training encoder, which is a pre-training BERT model, and a softmax classifier, which includes a "start position classifier" and an "end position classifier". The codes obtained by the encoder are classified by using a "softmax start position classifier", and the classification target aims to ensure that the classification probability of the start character belonging to the entity sequence is as high as possible, and the classification probability of the start character string not belonging to the entity sequence is as low as possible. The coding obtained by the coder is classified by using a softmax ending position classifier, and the classification target aims to ensure that the classification probability of the ending character belonging to the entity sequence is as high as possible, and the classification probability of the ending character string not belonging to the entity sequence is as low as possible. And finally, transmitting the encoder obtained by the entity boundary identification module to a third part.
Third part, named entity extraction module (FIG. 4)
And matching the knowledge base anchor text obtained by the first part with the item noun, marking the character string obtained by matching as an item name entity as input, and giving an input of 'identifying the item name in the text'. And splicing the two sections of input and inputting the two sections of input into a named entity extraction module. The main algorithms of the module are a pre-training encoder and a softmax classifier, wherein the pre-training encoder is initialized using the encoder obtained in the second part. The softmax classifier includes a "start position classifier" and an "end position classifier". The codes obtained by the encoder are classified by using a 'softmax starting position classifier', and the classification target aims to ensure that the classification probability of the starting character belonging to the entity sequence of the project name is as high as possible, and the classification probability of the starting character string not belonging to the entity sequence of the project name is as low as possible. The coding obtained by the coder is classified by using a softmax ending position classifier, and the classification target aims to ensure that the classification probability of ending characters belonging to the entity sequence of the project name is as high as possible, and the classification probability of ending character strings not belonging to the entity sequence of the project name is as low as possible. And re-predicting the input after each algorithm iteration is finished, and re-inputting the predicted result into the model. The iteration is stopped when the recognition rate of the model for the item name reaches the highest point on the verification set, and the encoder of the highest point round is transferred to the fourth part.
Fourth section, abbreviated project name identification Module (FIG. 5)
And inputting the expert labels into an abbreviated item name recognition module. The main algorithms of the module are a pre-training encoder and a softmax classifier, wherein the pre-training encoder is initialized using the encoder obtained in the third part. The softmax classifier includes a "start position classifier" and an "end position classifier". The codes obtained by the encoder are classified by using a "softmax start position classifier", and the classification target aims to ensure that the classification probability of the start character belonging to the abbreviated item name entity sequence is as high as possible, and the classification probability of the start character string not belonging to the abbreviated item name entity sequence is as low as possible. The codes obtained by the coder are classified by using a softmax ending position classifier, and the classification target aims to ensure that the classification probability of ending characters belonging to the entity sequence of the abbreviated item name is as high as possible, and the classification probability of ending character strings not belonging to the entity sequence of the abbreviated item name is as low as possible. And finally, packaging the encoder of the abbreviated item name recognition module into an abbreviated item name recognition model.
In practical application, only the test text is input into the abbreviated item name recognition model, so that the abbreviated item name in the test text can be recognized.
Experimental data
The recognition rate of the abbreviated project name recognition model obtained by the invention on the project name can reach 98.43%, and the recall rate can reach 97.41%. Compared with a common knowledge integration deep learning method BERT/MRC-NER, the recognition rate of 3.58%/1.46% and the recall rate of 2.07%/1.04% are improved based on the abbreviation project name recognition model integrating the structured entity information.
The above examples are provided for the purpose of describing the present invention only and are not intended to limit the scope of the present invention. The scope of the invention is defined by the appended claims. Various equivalents and modifications that do not depart from the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (6)
1. An abbreviated project name recognition method based on merging structured entity information comprises the following steps:
1) Marking entity sequences in the encyclopedia knowledge base to obtain knowledge base anchor texts, and constructing a common item name dictionary through collected common item names;
2) Inputting the knowledge base anchor text into a pre-training encoder, classifying the character strings output by the pre-training encoder according to the fact that the character string classification probability belonging to the entity sequence is as high as possible and the character string classification probability not belonging to the entity sequence is as low as possible, so as to train the pre-training encoder, and obtaining a first encoder;
3) Matching the text of the knowledge base anchor text with the common project names in the common project name dictionary, inputting the matching result into a first encoder, classifying the character strings output by the first encoder according to the character string classification probability belonging to the project name entity sequence as high as possible and the character string classification probability not belonging to the project name entity sequence as low as possible so as to train the first encoder and obtain a second encoder;
4) Inputting marked abbreviation item name recognition data into a second encoder, classifying the character strings output by the second encoder according to the fact that the character string classification probability belonging to the abbreviation item name entity sequence is as high as possible and the character string classification probability not belonging to the abbreviation item name entity sequence is as low as possible, so as to train the second encoder, and obtaining an abbreviation item name recognition model;
5) Inputting a test text into the abbreviated item name recognition model, and recognizing abbreviated item names in the test text;
wherein, by using a softmax classifier, the character string output by the pre-training encoder, the character string output by the first encoder and the character string output by the second encoder are classified; the softmax classifier included: a softmax start position classifier and a softmax end position classifier;
the softmax start position classifier and the softmax end position classifier classify the strings output by the pre-training encoder by the following strategy:
classifying the character strings output by the pre-training encoder according to the fact that the classification probability of the beginning characters belonging to the entity sequences is as high as possible and the classification probability of the beginning character strings not belonging to the entity sequences is as low as possible;
classifying the character strings output by the pre-training encoder according to the fact that the classification probability of the ending character belonging to the entity sequence is as high as possible and the classification probability of the ending character string not belonging to the entity sequence is as low as possible;
the softmax start position classifier and the softmax end position classifier classify the character string output by the first encoder by the following strategy:
classifying the character strings output by the first encoder by a softmax starting position classifier according to the fact that the classification probability of the starting character belonging to the entity sequence of the project name is as high as possible and the classification probability of the starting character string not belonging to the entity sequence of the project name is as low as possible;
classifying the character strings output by the first encoder according to the fact that the classification probability of the ending character belonging to the entity sequence of the project name is as high as possible and the classification probability of the ending character string not belonging to the entity sequence of the project name is as low as possible;
the softmax start position classifier and the softmax end position classifier classify the character string output by the second encoder by the following strategy:
classifying the character strings output by the second encoder according to the fact that the classification probability of the initial character belonging to the abbreviated item name entity sequence is as high as possible and the classification probability of the initial character string not belonging to the abbreviated item name entity sequence is as low as possible;
and classifying the character strings output by the second encoder according to the fact that the classification probability of the ending character belonging to the entity sequence of the abbreviated item name is as high as possible and the classification probability of the ending character string not belonging to the entity sequence of the abbreviated item name is as low as possible.
2. The method of claim 1, wherein the entity sequence is obtained by:
1) Constructing a regular expression;
2) And matching texts of the encyclopedia knowledge base through the regular expression to obtain an entity sequence.
3. The method of claim 2, wherein the format of the regular expression comprises: r' \ [ {2} (.
4. The method of claim 1, wherein the pre-training encoder is a pre-training BERT model.
5. A storage medium having a computer program stored therein, wherein the computer program is arranged to perform the method of any of claims 1-4 when run.
6. An electronic device comprising a memory, in which a computer program is stored, and a processor arranged to run the computer program to perform the method of any of claims 1-4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011481330.9A CN112651243B (en) | 2020-12-15 | 2020-12-15 | Abbreviated project name identification method based on integrated structured entity information and electronic device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011481330.9A CN112651243B (en) | 2020-12-15 | 2020-12-15 | Abbreviated project name identification method based on integrated structured entity information and electronic device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112651243A CN112651243A (en) | 2021-04-13 |
| CN112651243B true CN112651243B (en) | 2023-11-03 |
Family
ID=75354167
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011481330.9A Active CN112651243B (en) | 2020-12-15 | 2020-12-15 | Abbreviated project name identification method based on integrated structured entity information and electronic device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112651243B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10388274B1 (en) * | 2016-03-31 | 2019-08-20 | Amazon Technologies, Inc. | Confidence checking for speech processing and query answering |
| CN111078875A (en) * | 2019-12-03 | 2020-04-28 | 哈尔滨工程大学 | Method for extracting question-answer pairs from semi-structured document based on machine learning |
| CN111626063A (en) * | 2020-07-28 | 2020-09-04 | 浙江大学 | Text intention identification method and system based on projection gradient descent and label smoothing |
| CN112000791A (en) * | 2020-08-26 | 2020-11-27 | 哈电发电设备国家工程研究中心有限公司 | Motor fault knowledge extraction system and method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11899786B2 (en) * | 2019-04-15 | 2024-02-13 | Crowdstrike, Inc. | Detecting security-violation-associated event data |
-
2020
- 2020-12-15 CN CN202011481330.9A patent/CN112651243B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10388274B1 (en) * | 2016-03-31 | 2019-08-20 | Amazon Technologies, Inc. | Confidence checking for speech processing and query answering |
| CN111078875A (en) * | 2019-12-03 | 2020-04-28 | 哈尔滨工程大学 | Method for extracting question-answer pairs from semi-structured document based on machine learning |
| CN111626063A (en) * | 2020-07-28 | 2020-09-04 | 浙江大学 | Text intention identification method and system based on projection gradient descent and label smoothing |
| CN112000791A (en) * | 2020-08-26 | 2020-11-27 | 哈电发电设备国家工程研究中心有限公司 | Motor fault knowledge extraction system and method |
Non-Patent Citations (2)
| Title |
|---|
| 基于CNN和深层语义匹配的中文实体链接模型;吴晓崇;段跃兴;张月琴;闫雄;;计算机工程与科学(08);第187-193页 * |
| 基于深度学习的患者安全事件的命名实体识别;周亮杰;马敬东;;中华医学图书情报杂志(06);第5-10页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112651243A (en) | 2021-04-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113177124B (en) | Method and system for constructing knowledge graph in vertical field | |
| CN110334213B (en) | Method for identifying time sequence relation of Hanyue news events based on bidirectional cross attention mechanism | |
| CN113468888A (en) | Entity relation joint extraction method and device based on neural network | |
| CN109241330A (en) | The method, apparatus, equipment and medium of key phrase in audio for identification | |
| CN113946677B (en) | Event identification and classification method based on bidirectional cyclic neural network and attention mechanism | |
| CN112749562A (en) | Named entity identification method, device, storage medium and electronic equipment | |
| CN113191148A (en) | Rail transit entity identification method based on semi-supervised learning and clustering | |
| CN116416480B (en) | Visual classification method and device based on multi-template prompt learning | |
| CN116127090B (en) | Aviation system knowledge graph construction method based on fusion and semi-supervision information extraction | |
| CN116956929B (en) | Multi-feature fusion named entity recognition method and device for bridge management text data | |
| CN114881043B (en) | Deep learning model-based legal document semantic similarity evaluation method and system | |
| CN113051887A (en) | Method, system and device for extracting announcement information elements | |
| CN110866172B (en) | Data analysis method for block chain system | |
| CN116089610A (en) | Label identification method and device based on industry knowledge | |
| CN113901813A (en) | Event extraction method based on topic features and implicit sentence structure | |
| CN112148879B (en) | Computer readable storage medium for automatically labeling code with data structure | |
| CN112651243B (en) | Abbreviated project name identification method based on integrated structured entity information and electronic device | |
| CN115408506B (en) | NL2SQL method combining semantic analysis and semantic component matching | |
| CN116595023A (en) | Address information updating method and device, electronic equipment and storage medium | |
| CN116451691A (en) | Small sample named entity identification method for entity hierarchy information enhanced prototype characterization | |
| CN114595338A (en) | Entity relation joint extraction system and method based on mixed feature representation | |
| CN114677526A (en) | Image classification method, device, equipment and medium | |
| CN112015891A (en) | Method and system for classifying messages of network inquiry platform based on deep neural network | |
| CN110543560A (en) | Long text classification and identification method, device and medium based on convolutional neural network | |
| CN113313184B (en) | Heterogeneous integrated self-bearing technology liability automatic detection method |
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 |