CN113705728B - Classification and classification list intelligent marking method - Google Patents
Classification and classification list intelligent marking method Download PDFInfo
- Publication number
- CN113705728B CN113705728B CN202111102610.9A CN202111102610A CN113705728B CN 113705728 B CN113705728 B CN 113705728B CN 202111102610 A CN202111102610 A CN 202111102610A CN 113705728 B CN113705728 B CN 113705728B
- Authority
- CN
- China
- Prior art keywords
- text
- match
- label
- max
- field
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- 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
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an intelligent marking method for a classified and graded list, relates to the technical field of communication, and solves the technical problems that at present, under the large data magnitude, the marking of a data list is time-consuming and labor-consuming, and the marking quality is low. According to the technical scheme, table and field information are read through different databases, text line formats are unified, then preprocessing, chinese-English word segmentation and English translation operations are carried out, text rough classification is firstly carried out based on content diversity and complexity of the text lines, text lines with marked overall results are obtained, and training and short text classification are further carried out based on fastText models. The invention has high-speed and high-quality models in short text classification, trains and tests the text lines to obtain reasonable marking results of the text lines, and performs intelligent marking, thereby saving time and labor.
Description
Technical Field
The invention relates to the technical field of communication, in particular to an intelligent marking method for a classified and graded list.
Background
Short text classification has been widely used in scenes such as public opinion classification, news classification, etc.; in the field of data security, classification and grading of data lists encounters a huge bottleneck in implementation; the data list is generally composed of databases with different storage modes, tables with different naming forms and field information, and the orders of magnitude are tens of thousands to millions; in the face of such magnitude of table and field information, it is an obstacle to classifying the marking, which presents challenges to the number and expertise of the marking personnel, and it takes days to months; at the same time, the quality of the marking can also become an important issue.
Disclosure of Invention
The invention aims to provide an intelligent marking method for a classification and grading list, which is used for classifying and grading by classifying short texts, extracting keywords of the short texts, obtaining vectorized short texts, clustering the short texts and classifying the short texts, is intelligent and reasonable in marking, saves time and labor and improves marking quality.
In order to achieve the above object, the present invention provides the following technical solutions: an intelligent marking method for a classified and graded list comprises the following steps:
s1, reading tables and field information of different databases, and processing the tables and the field information into text lines with uniform formats, wherein the contents of the text lines comprise field names, field notes, table names and table notes; simultaneously reading all the tags; preprocessing a text line, including deactivating words and punctuation marks; then, performing Chinese and English word segmentation on the text line, and performing Chinese word segmentation on the tag; the English after word segmentation is translated into corresponding Chinese through a word stock in English translation; at this time, a TEXT line of the word segmentation is obtained, the TEXT is recorded, the TEXT line is divided into FIELD information and TABLE information according to the content of the TEXT line, and the text_field and the text_table are recorded respectively; the LABEL of the word is divided into a LABEL and a background LABEL according to whether the specific field is the specific field, and LABEL and label_BG are respectively recorded; and assuming that the number of tags is class_num; so far, the text line and the label are segmented;
s2, matching text and labels; traversing the TEXT, traversing the LABEL for the text_field of each TEXT, and recording the TEXT FIELD information and the number of words matched with the LABEL to obtain a list with the length of CLASS_NUM; the maximum value in the record list is MATCH_MAX; according to whether MATCH_MAX is unique, the following 2 cases are classified:
s2.1. The MATCH_MAX is unique, and the label of the TEXT coarse classification is the label corresponding to the MATCH_MAX;
s2.2.MATCH_MAX is not unique, all tags with the matching number of MATCH_MAX are obtained, MATCH_MAX_LABEL is recorded, the number of the tags is MATCH_MAX_LABEL_NUM, TEXT_TABLE of TEXT is traversed, the number of words matched with TEXT TABLE information and background tags is recorded, and a list with the length of MATCH_MAX_LABEL_NUM is obtained; the maximum value in the record list is MATCH_TABLE_MAX; according to whether MATCH_TABLE_MAX is unique, the following 2 cases are also classified:
s2.2.1.match_table_max is unique, and the tag of the TEXT coarse classification is the tag corresponding to match_table_max;
s2.2.2.MATCH_TABLE_MAX is not unique, all the LABELs with the matching number of MATCH_TABLE_MAX in S2.2 are obtained, MATCH_TABLE_MAX_LABEL is recorded, TEXT_FIELD of TEXT is traversed, LABEL with the matching number of FIELDs and the proportion of the matching number to the total number of LABEL are recorded, MATCH_CHAR and MATCH_CHAR_RATIO are respectively recorded, the VALUEs of MATCH_CHAR+MATCH_CHAR_RATIO are recorded as matching VALUEs MATCH_VALUE, a list of matching VALUEs is obtained, and the LABEL corresponding to the highest matching VALUE is selected as a rough-classification LABEL; so far, according to the field information matching number, the table information matching number and the priority of the matching word number, the rough classification of most text lines is completed;
s3, according to actual conditions, the situation that the number of text lines corresponding to some labels in the S2 text rough classification is extremely small is dealt with, and a small amount of data expansion is carried out on the text lines;
s4, training a model based on fastText and classifying short texts; for the short text in S3, converting the text line into a format according to the read-in format of the fastText algorithm, such as index conversion of labels; and setting a THRESHOLD of confidence, and marking a text line as a corresponding label when the confidence exceeds 0.9, for example, by threshold=0.9; and traversing TEXT, and marking the TEXT according to the comparison of the confidence result and the THRESHOLD.
In step S4, the fastText-based model includes an N-gram and a hierarchical softmax.
The invention describes an intelligent marking method for a classified and graded list, wherein the text of the classified and graded list is usually composed of tables and fields with different rules and naming modes, the marking rate of rule strategy matching is low, and meanwhile, the misjudgment is higher; since text quality is usually not high, clustering results are also not ideal; the text is roughly classified by using the text-label matching method as a standard classification method: performing rough classification on the text lines according to the following priorities, wherein the number of field information matching words is greater than the number of table information matching words and greater than the number of field information matching words; the coarse classification result is subjected to fastText learning and testing, so that the marking rate is greatly improved compared with the prior art, and the misjudgment rate is also remarkably reduced; the matching words in the text rough classification are obtained by a test mode, namely experience; part of LABELs have the problems of overlong length, redundant content and the like, so that the LABELs are often provided with a plurality of words of other LABELs at the same time, which can obstruct the normal rough classification process, and therefore, the misjudgment bottleneck in rough classification of texts is solved by taking the weighted sum of the number of words matched by a field and the proportion of the number of matched words to the total number of LABEL as the classification standard; and the intelligent marking is realized, so that time and labor are saved.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the present invention, and other drawings may be obtained according to these drawings for a person having ordinary skill in the art.
FIG. 1 is a flow chart of the present invention.
Detailed Description
In order to make the technical scheme of the present invention better understood by those skilled in the art, the present invention will be further described in detail with reference to the accompanying drawings.
The invention provides an intelligent marking method for a classified and graded list as shown in figure 1,
1. firstly, reading tables and field information of different databases, and processing the tables and the field information into text lines with uniform formats, wherein the contents of the text lines comprise field names, field notes, table names and table notes 4 parts; simultaneously reading all the tags; preprocessing a text line, including deactivating words and punctuation marks; then, performing Chinese and English word segmentation on the text line, and performing Chinese word segmentation on the tag; the English after word segmentation is translated into corresponding Chinese through a word stock in English translation; at this time, a TEXT line of the word segmentation is obtained, the TEXT is recorded, the TEXT line is divided into FIELD information and TABLE information according to the content of the TEXT line, and the text_field and the text_table are recorded respectively; the LABEL of the word is divided into a LABEL and a background LABEL according to whether the specific field is the specific field, and LABEL and label_BG are respectively recorded; and assuming that the number of tags is class_num; so far, the text line and the label are segmented;
2. secondly, text coarse classification: text-tag matching; traversing the TEXT, traversing the LABEL for the text_field of each TEXT, and recording the TEXT FIELD information and the number of words matched with the LABEL to obtain a list with the length of CLASS_NUM; the maximum value in the record list is MATCH_MAX; according to whether MATCH_MAX is unique, the following 2 cases are classified; MATCH_MAX is unique, and the label of the TEXT coarse classification is the label corresponding to MATCH_MAX; acquiring all LABELs with matching number of MATCH_MAX, recording MATCH_MAX_LABEL, recording the number of the LABELs, recording MATCH_MAX_LABEL_NUM, traversing TEXT_TABLE in TEXT_MAX_LABEL, recording TEXT TABLE information and the number of words matched with background LABELs, and obtaining a list with length of MATCH_MAX_LABEL_NUM; the maximum value in the record list is MATCH_TABLE_MAX; according to whether MATCH_TABLE_MAX is unique, the following 2 cases are also classified; the only one MATCH_TABLE_MAX, the label of the TEXT rough classification is the label corresponding to MATCH_TABLE_MAX; acquiring all LABELs with the matching number of MATCH_TABLE_MAX in 2.2, recording MATCH_TABLE_MAX_LABEL, traversing LABEL of TEXT_TABLE_MAX_LABEL, recording the word number of FIELD matching and the proportion of the matching word number to the total word number of LABEL, recording the VALUEs of MATCH_CHAR and MATCH_CHAR_RATIO respectively, recording the VALUEs of MATCH_CHAR+MATCH_CHAR_RATIO as matching VALUEs, obtaining a list of matching VALUEs, and selecting the LABEL corresponding to the highest matching VALUE as a rough-classified LABEL; so far, according to the field information matching number, the table information matching number and the priority of the matching word number, the rough classification of most text lines is completed;
3. according to the actual situation, a small amount of data augmentation can be performed on text lines corresponding to some labels in the text rough classification under the condition that the number of the text lines is very small;
4. again, fastText-based model training and short text classification; fastttext is a model with high speed and high precision in text classification; the model is simple in structure, and N-gram and layering softmax are used in the implementation process, wherein the accuracy is improved, and the speed is improved; on the premise that the accuracy and the deep learning are quite equivalent, the training and testing speed is higher than that of the training and testing by several orders of magnitude; for the short text in 3, converting the text line into a format according to the reading format of the fastText algorithm, such as index conversion of labels; and setting a THRESHOLD of confidence, and marking a text line as a corresponding label when the confidence exceeds 0.9, for example, by threshold=0.9; traversing TEXT, and marking a TEXT according to the comparison of the confidence result and the THRESHOLD;
5. so far, the invention completes the whole process from reading the table and the field information to marking; the technical problem that marking of the classified list is time-consuming and low-efficient is solved.
While certain exemplary embodiments of the present invention have been described above by way of illustration only, it will be apparent to those of ordinary skill in the art that modifications may be made to the described embodiments in various different ways without departing from the spirit and scope of the invention. Accordingly, the drawings and description are to be regarded as illustrative in nature and not as restrictive of the scope of the invention, which is defined by the appended claims.
Claims (2)
1. An intelligent marking method for classified and graded lists is characterized in that,
s1, reading tables and field information of different databases, and processing the tables and the field information into text lines with uniform formats, wherein the contents of the text lines comprise field names, field notes, table names and table notes; simultaneously reading all the tags; preprocessing a text line, including deactivating words and punctuation marks; then, performing Chinese and English word segmentation on the text line, and performing Chinese word segmentation on the tag; the English after word segmentation is translated into corresponding Chinese through a word stock in English translation; at this time, a TEXT line of the word segmentation is obtained, the TEXT is recorded, the TEXT line is divided into FIELD information and TABLE information according to the content of the TEXT line, and the text_field and the text_table are recorded respectively; the LABEL of the word is divided into a LABEL and a background LABEL according to whether the specific field is the specific field, and LABEL and label_BG are respectively recorded; and assuming that the number of tags is class_num; so far, the text line and the label are segmented;
s2, matching text and labels; traversing the TEXT, traversing the LABEL for the text_field of each TEXT, and recording the TEXT FIELD information and the number of words matched with the LABEL to obtain a list with the length of CLASS_NUM; the maximum value in the record list is MATCH_MAX; according to whether MATCH_MAX is unique, the following 2 cases are classified:
s2.1, the MATCH_MAX is unique, and the label of the TEXT rough classification is the label corresponding to the MATCH_MAX;
s2.2, acquiring all LABELs with matching number of MATCH_MAX, recording MATCH_MAX_LABEL, recording the number of the LABELs, recording MATCH_MAX_LABEL_NUM, traversing TEXT_TABLE in TEXT_MAX_LABEL, recording the number of words matched with TEXT TABLE information and background LABELs, and obtaining a list with length of MATCH_MAX_LABEL_NUM; the maximum value in the record list is MATCH_TABLE_MAX; according to whether MATCH_TABLE_MAX is unique, the following 2 cases are also classified:
s2.2.1 the MATCH_TABLE_MAX is unique, and the tags of the TEXT coarse classification are the tags corresponding to MATCH_TABLE_MAX;
s2.2.2 all the LABELs with MATCH number MATCH_TABLE_MAX in S2.2 are obtained, MATCH_TABLE_MAX_LABEL is recorded, TEXT_FIELD of TEXT is traversed, LABEL of MATCH_TABLE_MAX_LABEL is traversed, the 'number of words matched with FIELDs' and the 'proportion of the number of matching words to the total number of LABEL' are recorded, MATCH_CHAR and MATCH_CHAR_RATIO are recorded, the VALUEs of MATCH_CHAR+MATCH_CHAR_RATIO are recorded as MATCH VALUEs MATCH_VALUE, a list of MATCH VALUEs is obtained, and the LABEL corresponding to the highest MATCH VALUE is selected as the LABEL of rough classification; so far, according to the field information matching number, the table information matching number and the priority of the matching word number, the rough classification of most text lines is completed;
s3, according to actual conditions, dealing with the situation that the number of text lines corresponding to some labels in the S2 text rough classification is extremely small, and carrying out small data augmentation on the text lines;
s4, training a model based on fastText and classifying short texts; for the short text in S3, converting the text line according to the read-in format of the fastText algorithm; setting a THRESHOLD of the confidence, and recording THRESHOLD; and traversing TEXT, and marking the TEXT according to the comparison of the confidence result and the THRESHOLD.
2. The intelligent marking method for the classified list according to claim 1, wherein: in step S4, the fastText-based model includes an N-gram and a hierarchical softmax.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111102610.9A CN113705728B (en) | 2021-09-18 | 2021-09-18 | Classification and classification list intelligent marking method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111102610.9A CN113705728B (en) | 2021-09-18 | 2021-09-18 | Classification and classification list intelligent marking method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113705728A CN113705728A (en) | 2021-11-26 |
CN113705728B true CN113705728B (en) | 2023-08-01 |
Family
ID=78661364
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111102610.9A Active CN113705728B (en) | 2021-09-18 | 2021-09-18 | Classification and classification list intelligent marking method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113705728B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109522298A (en) * | 2018-08-29 | 2019-03-26 | 云南电网有限责任公司信息中心 | Data cleaning method for CIM |
CN109766410A (en) * | 2019-01-07 | 2019-05-17 | 东华大学 | A kind of newsletter archive automatic classification system based on fastText algorithm |
CN110059181A (en) * | 2019-03-18 | 2019-07-26 | 中国科学院自动化研究所 | Short text stamp methods, system, device towards extensive classification system |
WO2020215457A1 (en) * | 2019-04-26 | 2020-10-29 | 网宿科技股份有限公司 | Adversarial learning-based text annotation method and device |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7783637B2 (en) * | 2003-09-30 | 2010-08-24 | Microsoft Corporation | Label system-translation of text and multi-language support at runtime and design |
-
2021
- 2021-09-18 CN CN202111102610.9A patent/CN113705728B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109522298A (en) * | 2018-08-29 | 2019-03-26 | 云南电网有限责任公司信息中心 | Data cleaning method for CIM |
CN109766410A (en) * | 2019-01-07 | 2019-05-17 | 东华大学 | A kind of newsletter archive automatic classification system based on fastText algorithm |
CN110059181A (en) * | 2019-03-18 | 2019-07-26 | 中国科学院自动化研究所 | Short text stamp methods, system, device towards extensive classification system |
WO2020215457A1 (en) * | 2019-04-26 | 2020-10-29 | 网宿科技股份有限公司 | Adversarial learning-based text annotation method and device |
Also Published As
Publication number | Publication date |
---|---|
CN113705728A (en) | 2021-11-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107766371B (en) | Text information classification method and device | |
CN112131920B (en) | Data structure generation for table information in scanned images | |
CN111209728B (en) | Automatic labeling and inputting method for test questions | |
CN111061882A (en) | Knowledge graph construction method | |
CN112036184A (en) | Entity identification method, device, computer device and storage medium based on BilSTM network model and CRF model | |
CN111753514B (en) | Automatic generation method and device of patent application text | |
CN112395858B (en) | Multi-knowledge point labeling method and system integrating test question data and answer data | |
CN115034218A (en) | Chinese grammar error diagnosis method based on multi-stage training and editing level voting | |
CN114090736A (en) | Enterprise industry identification system and method based on text similarity | |
Lehenmeier et al. | Layout detection and table recognition–recent challenges in digitizing historical documents and handwritten tabular data | |
CN115292490A (en) | Analysis algorithm for policy interpretation semantics | |
CN111178080A (en) | Named entity identification method and system based on structured information | |
CN112632282B (en) | Chinese and English thesis data classification and query method | |
CN117034948B (en) | Paragraph identification method, system and storage medium based on multi-feature self-adaptive fusion | |
CN109685103A (en) | A kind of text Multi-label learning method based on broad sense K mean algorithm | |
CN113705728B (en) | Classification and classification list intelligent marking method | |
CN112036330A (en) | Text recognition method, text recognition device and readable storage medium | |
CN116304064A (en) | Text classification method based on extraction | |
CN110941703A (en) | Integrated resume information extraction method based on machine learning and fuzzy rules | |
CN116226371A (en) | Digital economic patent classification method | |
TW200409046A (en) | Optical character recognition device, document searching system, and document searching program | |
CN111400606B (en) | Multi-label classification method based on global and local information extraction | |
CN114419645A (en) | Contract intelligent analysis method based on AI | |
CN114328902A (en) | Text labeling model construction method and device | |
CN114492419B (en) | Text labeling method, system and device based on newly added key words in labeling |
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 |