CN116227460A - Address element analysis method and device based on semantic expression - Google Patents

Address element analysis method and device based on semantic expression Download PDF

Info

Publication number
CN116227460A
CN116227460A CN202211557672.3A CN202211557672A CN116227460A CN 116227460 A CN116227460 A CN 116227460A CN 202211557672 A CN202211557672 A CN 202211557672A CN 116227460 A CN116227460 A CN 116227460A
Authority
CN
China
Prior art keywords
data
address
analysis
result
model
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.)
Pending
Application number
CN202211557672.3A
Other languages
Chinese (zh)
Inventor
陶闯
裘靖宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Weizhi Zhuoxin Information Technology Co ltd
Original Assignee
Shanghai Weizhi Zhuoxin Information Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shanghai Weizhi Zhuoxin Information Technology Co ltd filed Critical Shanghai Weizhi Zhuoxin Information Technology Co ltd
Priority to CN202211557672.3A priority Critical patent/CN116227460A/en
Publication of CN116227460A publication Critical patent/CN116227460A/en
Pending legal-status Critical Current

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
    • G06F40/205—Parsing
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses an address element analysis method and device based on semantic expression, wherein the method comprises the following steps: when detecting that element analysis operation is required, determining standard address data corresponding to the address data; the standard address data is address data expressed in a set text format after data preprocessing; inputting the standard address data into the trained address element analysis model to obtain an address element analysis result; the address element analysis result is an analysis result obtained after the character division operation is performed on the basis of the determined predicted address element category corresponding to the standard address data. Therefore, the invention can adopt the trained address element analysis model to realize element analysis of the address data, improve the determination accuracy and efficiency of the address element analysis result, and further improve the analysis accuracy and efficiency of the address element, thereby being beneficial to better solving the address service requirement based on the address element and improving the use experience of users aiming at the address element related functions and applications.

Description

Address element analysis method and device based on semantic expression
Technical Field
The invention relates to the technical field of address semantic parsing, in particular to an address element parsing method and device based on semantic expression.
Background
Along with the arrival and rapid development of the Internet of things big data age, data analysis application is gradually and widely and frequently carried out, wherein address data has the characteristics of strong knowledge, multiple ambiguous place names, frequent data transition and the like, and address element analysis is used as an important component of address coding, so that the demand is high and the application is wide.
Currently, aiming at the address element analysis mode, the method mainly based on statistical information or the method based on rules is used for realizing the address element analysis by modeling a word sequence information training model among words, ignoring the semantic information of the words, and has poor recognition effect, so that the requirement of the current address service is difficult to meet; the method based on rules realizes address element analysis by matching feature words, constructing a place name dictionary and the like, and has high matching speed but is difficult to solve the naming diversity of the address elements. It can be seen that both of the above-mentioned address element resolution methods have the problem of low address element resolution accuracy. Therefore, it is important to provide a new address element analysis method to improve the resolution accuracy of the address element.
Disclosure of Invention
The invention aims to solve the technical problem of providing an address element analysis method and an address element analysis device based on semantic expression, which can improve the analysis accuracy of the address element.
In order to solve the technical problem, the first aspect of the present invention discloses an address element parsing method based on semantic expression, which comprises the following steps:
when detecting that address data need to be subjected to element analysis operation, determining standard address data corresponding to the address data; the standard address data are address data expressed in a set text format after data preprocessing;
inputting the standard address data into a trained address element analysis model to obtain a corresponding address element analysis result; and the address element analysis result is an analysis result which is obtained after the character division operation is performed on the basis of the determined predicted address element category and corresponds to the standard address data.
As an optional implementation manner, in the first aspect of the present invention, before the inputting the canonical address data into the trained address element analysis model to obtain a corresponding address element analysis result, the method further includes:
acquiring training set data for address element analysis training, and constructing and training an address element analysis model based on the training set data to obtain the address element analysis model after training;
And constructing and training an address element analysis model based on the training set data to obtain the trained address element analysis model, wherein the method comprises the following steps:
constructing a basic analysis model according to the semantic analysis module, the hidden information acquisition module and the element constraint module;
training the basic analytic model according to the training set data to obtain a trained analytic model;
according to the address prediction result of the address prediction of the training set data by the basic analysis model and the labeled address information corresponding to the training set data, calculating a model training loss value;
judging whether the model training loss value is smaller than or equal to a set model training loss value threshold value or not;
when the judgment result is yes, determining the trained analytic model as a trained address element analytic model;
and when the judgment result is negative, adjusting model training parameters, and executing model training operation based on the adjusted model training parameters until the model training loss value is smaller than or equal to the model training loss value threshold and/or the training times reach a preset time threshold.
In a first aspect of the present invention, training the basic analytical model according to the training set data to obtain a trained analytical model includes:
For any data to be trained in the training set data, inputting the data to be trained into the corresponding semantic analysis module to obtain a word vector representation result corresponding to characters in the data to be trained;
inputting the word vector representation result into the corresponding hidden information acquisition module to obtain a hidden representation result corresponding to the character, wherein the hidden representation result comprises a forward hidden semantic result and/or a reverse hidden semantic result, and determining front and rear semantic information corresponding to the character according to the hidden representation result and data obtained by performing data preprocessing operation on the data to be trained;
and determining a corresponding address prediction result according to the element constraint module, the front and rear semantic information and the data to be trained, and taking a basic analysis model after outputting the address prediction result as a trained analysis model.
In an optional implementation manner, in a first aspect of the present invention, the determining, according to the element constraint module, the front-back semantic information, and the data to be trained, a corresponding address prediction result includes:
inputting the front and rear semantic information into the element constraint module to obtain at least one sequence tag prediction result corresponding to the characters in the data to be trained, and calculating to obtain a first validity corresponding to each sequence tag prediction result by combining a tag analysis algorithm;
Determining a target sequence tag prediction result from all the sequence tag prediction results according to the first validity and the set tag analysis conditions;
performing normalization processing operation on the first validity corresponding to each prediction tag in the target sequence tag prediction result, determining the second validity of each prediction tag corresponding to the character, and determining the target prediction tag of the character according to the second validity of all the prediction tags;
and determining an address prediction result corresponding to the data to be trained according to target prediction labels of all the characters corresponding to the data to be trained.
In an optional implementation manner, in a first aspect of the present invention, the determining, according to the hidden representation result and data obtained by performing a data preprocessing operation on the data to be trained, front and rear semantic information corresponding to the character includes:
when the hidden representation result comprises the forward hidden semantic result and the reverse hidden semantic result, determining corresponding forward input text and reverse input text according to data obtained by performing data preprocessing operation on the data to be trained;
Respectively inputting the forward input text and the reverse input text into corresponding gating cyclic neural networks to obtain a processing result;
and determining front and rear semantic information corresponding to the character according to the processing result, the forward latent semantic result, the reverse latent semantic result and the set latent semantic processing conditions.
In a first aspect of the present invention, as an optional implementation manner, the calculating a model training loss value according to an address prediction result of address prediction for the training set data by the base parsing model and labeled address information corresponding to the training set data includes:
determining an analysis validity result corresponding to each sub-parameter in the analysis judgment parameters according to an address prediction result of address prediction of the training set data by the basic analysis model, labeling address information corresponding to the training set data and the set analysis judgment parameters; the analysis judgment parameters comprise one or more of accuracy parameters, recall rate parameters, stability parameters, analysis efficiency parameters and analysis cost performance parameters;
and determining the analysis validity conditions corresponding to the basic analysis model according to the set parameter weighting conditions and the analysis validity results corresponding to all the sub-parameters, and determining the corresponding model training loss values according to the analysis validity conditions.
As an optional implementation manner, in the first aspect of the present invention, the specific manner of performing the data preprocessing operation on the data to be trained is:
determining first processing data according to the data to be trained and the data cleaning conditions;
determining second processing data as data of which the data preprocessing operation is completed corresponding to the data to be trained according to the first processing data and the data format preprocessing conditions;
the data cleaning conditions comprise one or more of full-angle and half-angle conversion, unified bracket form, punctuation mark deletion, space deletion, foreign language letter case conversion, chinese character simplified conversion, unified language form and unified character form.
The second aspect of the invention discloses an address element analysis device based on semantic expression, which comprises:
the determining module is used for determining the standard address data corresponding to the address data when detecting that the address data need to be subjected to element analysis operation; the standard address data are address data expressed in a set text format after data preprocessing;
the element analysis module is used for inputting the standard address data into the trained address element analysis model to obtain a corresponding address element analysis result; and the address element analysis result is an analysis result which is obtained after the character division operation is performed on the basis of the determined predicted address element category and corresponds to the standard address data.
As an alternative embodiment, in the second aspect of the present invention, the apparatus further includes:
the information acquisition module is used for acquiring training set data for address element analysis training before the element analysis module inputs the standard address data into the trained address element analysis model to obtain a corresponding address element analysis result;
the model training module is used for constructing and training an address element analysis model based on the training set data to obtain the address element analysis model after training;
and the model training module builds and trains an address element analysis model based on the training set data, and the mode for obtaining the trained address element analysis model specifically comprises the following steps:
constructing a basic analysis model according to the semantic analysis module, the hidden information acquisition module and the element constraint module;
training the basic analytic model according to the training set data to obtain a trained analytic model;
according to the address prediction result of the address prediction of the training set data by the basic analysis model and the labeled address information corresponding to the training set data, calculating a model training loss value;
Judging whether the model training loss value is smaller than or equal to a set model training loss value threshold value or not;
when the judgment result is yes, determining the trained analytic model as a trained address element analytic model;
and when the judgment result is negative, adjusting model training parameters, and executing model training operation based on the adjusted model training parameters until the model training loss value is smaller than or equal to the model training loss value threshold and/or the training times reach a preset time threshold.
In a second aspect of the present invention, the model training module trains the basic analytical model according to the training set data, and the method for obtaining the trained analytical model specifically includes:
for any data to be trained in the training set data, inputting the data to be trained into the corresponding semantic analysis module to obtain a word vector representation result corresponding to characters in the data to be trained;
inputting the word vector representation result into the corresponding hidden information acquisition module to obtain a hidden representation result corresponding to the character, wherein the hidden representation result comprises a forward hidden semantic result and/or a reverse hidden semantic result, and determining front and rear semantic information corresponding to the character according to the hidden representation result and data obtained by performing data preprocessing operation on the data to be trained;
And determining a corresponding address prediction result according to the element constraint module, the front and rear semantic information and the data to be trained, and taking a basic analysis model after outputting the address prediction result as a trained analysis model.
In a second aspect of the present invention, the mode of determining, by the model training module, the corresponding address prediction result according to the element constraint module, the front-back semantic information, and the data to be trained specifically includes:
inputting the front and rear semantic information into the element constraint module to obtain at least one sequence tag prediction result corresponding to the characters in the data to be trained, and calculating to obtain a first validity corresponding to each sequence tag prediction result by combining a tag analysis algorithm;
determining a target sequence tag prediction result from all the sequence tag prediction results according to the first validity and the set tag analysis conditions;
performing normalization processing operation on the first validity corresponding to each prediction tag in the target sequence tag prediction result, determining the second validity of each prediction tag corresponding to the character, and determining the target prediction tag of the character according to the second validity of all the prediction tags;
And determining an address prediction result corresponding to the data to be trained according to target prediction labels of all the characters corresponding to the data to be trained.
In a second aspect of the present invention, the mode of determining the front and rear semantic information corresponding to the character by the model training module according to the hidden representation result and the data obtained by performing the data preprocessing operation on the data to be trained specifically includes:
when the hidden representation result comprises the forward hidden semantic result and the reverse hidden semantic result, determining corresponding forward input text and reverse input text according to data obtained by performing data preprocessing operation on the data to be trained;
respectively inputting the forward input text and the reverse input text into corresponding gating cyclic neural networks to obtain a processing result;
and determining front and rear semantic information corresponding to the character according to the processing result, the forward latent semantic result, the reverse latent semantic result and the set latent semantic processing conditions.
In a second aspect of the present invention, as an optional implementation manner, the mode of calculating the model training loss value by the model training module according to the address prediction result of the address prediction of the basic analysis model on the training set data and the labeled address information corresponding to the training set data specifically includes:
Determining an analysis validity result corresponding to each sub-parameter in the analysis judgment parameters according to an address prediction result of address prediction of the training set data by the basic analysis model, labeling address information corresponding to the training set data and the set analysis judgment parameters; the analysis judgment parameters comprise one or more of accuracy parameters, recall rate parameters, stability parameters, analysis efficiency parameters and analysis cost performance parameters;
and determining the analysis validity conditions corresponding to the basic analysis model according to the set parameter weighting conditions and the analysis validity results corresponding to all the sub-parameters, and determining the corresponding model training loss values according to the analysis validity conditions.
As an optional implementation manner, in the second aspect of the present invention, the specific manner of performing the data preprocessing operation on the data to be trained is:
determining first processing data according to the data to be trained and the data cleaning conditions;
determining second processing data as data of which the data preprocessing operation is completed corresponding to the data to be trained according to the first processing data and the data format preprocessing conditions;
The data cleaning conditions comprise one or more of full-angle and half-angle conversion, unified bracket form, punctuation mark deletion, space deletion, foreign language letter case conversion, chinese character simplified conversion, unified language form and unified character form.
The third aspect of the present invention discloses another address element parsing device based on semantic expression, the device comprising:
a memory storing executable program code;
a processor coupled to the memory;
the processor calls the executable program code stored in the memory to execute the address element analysis method based on semantic expression disclosed in the first aspect of the invention.
A fourth aspect of the present invention discloses a computer storage medium storing computer instructions for executing the address element parsing method based on semantic expression disclosed in the first aspect of the present invention when the computer instructions are called.
Compared with the prior art, the embodiment of the invention has the following beneficial effects:
in the embodiment of the invention, when detecting that the address data needs to be subjected to element analysis operation, determining the standard address data corresponding to the address data; the standard address data is address data expressed in a set text format after data preprocessing; inputting the standard address data into a trained address element analysis model to obtain a corresponding address element analysis result; the address element analysis result is an analysis result obtained after the character division operation is performed on the basis of the determined predicted address element category corresponding to the standard address data. Therefore, the invention can adopt the trained address element analysis model to realize element analysis of the address data, improves the running stability and reliability of the trained address element analysis model, is beneficial to improving the determination accuracy and efficiency of the address element analysis result, and further improves the analysis accuracy and efficiency of the address element, thereby being beneficial to better solving the address service requirement based on the address element and improving the use experience of users aiming at the address element related functions and applications.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of an address element parsing method based on semantic expression according to an embodiment of the present invention;
FIG. 2 is a flow chart of another semantic expression-based address element parsing method according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of an address element parsing device based on semantic expression according to an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of another address element parsing device based on semantic expression according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of an address element parsing apparatus based on semantic expression according to an embodiment of the present invention;
fig. 6 is a parsing training flow chart and a parsing application flow chart corresponding to an address element parsing device based on semantic expression according to an embodiment of the present invention.
Detailed Description
In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The terms first, second and the like in the description and in the claims and in the above-described figures are used for distinguishing between different objects and not necessarily for describing a sequential or chronological order. Furthermore, the terms "comprise" and "have," as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, apparatus, article, or article that comprises a list of steps or elements is not limited to only those listed but may optionally include other steps or elements not listed or inherent to such process, method, article, or article.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the invention. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.
The invention discloses an address element analysis method and device based on semantic expression, which can adopt a trained address element analysis model to realize element analysis of address data, improve the running stability and reliability of the trained address element analysis model, be favorable for improving the determination accuracy and efficiency of an address element analysis result, and further improve the analysis accuracy and efficiency of the address element, thereby being favorable for better solving the address service requirement based on the address element and improving the use experience of a user aiming at the address element related function and application. The following will describe in detail.
Example 1
Referring to fig. 1, fig. 1 is a flow chart of an address element parsing method based on semantic expression according to an embodiment of the present invention. The method described in fig. 1 may be applied to an address element parsing apparatus based on semantic expression, where the apparatus may include a server, where the server includes a local server or a cloud server, and embodiments of the present invention are not limited. As shown in fig. 1, the address element parsing method based on semantic expression includes the following operations:
101. When detecting that the address data needs to be subjected to element analysis operation, determining the standard address data corresponding to the address data; the standard address data is address data expressed in a set text format after data preprocessing.
Alternatively, the set text format may be any format capable of normalizing the text character elements of the address, such as BIES format with words as units, which is not limited by the embodiment of the present invention. For example, BIES format specifically represents: b (Begin) indicates the beginning of the word as an address element, I (In) indicates the middle of the word as an address element, E (End) indicates the End of the word as an address element, and S (Single) indicates the word as a separate address element.
For example, the original address data is denoted as "Shanghai city (city) Pudong New region (distribution) Zhang Jiangzhen (township) in Koch (road) torch core research and development building (poi)", and the BIES format corresponding to the original address data may be denoted as "Shanghai (B-city) sea (I-city) E-city) Pu (B-distribution) east (I-distribution) New (I-distribution) region (E-distribution) E-Township". The description is omitted herein.
102. Inputting the standard address data into the trained address element analysis model to obtain a corresponding address element analysis result; the address element analysis result is an analysis result obtained after the character division operation is performed on the basis of the determined predicted address element category corresponding to the standard address data.
Therefore, the address element analysis method based on semantic expression described by the embodiment of the invention can adopt the trained address element analysis model to realize element analysis of the address data, improve the running stability and reliability of the trained address element analysis model, be favorable for improving the determination accuracy and efficiency of the address element analysis result, and further improve the analysis accuracy and efficiency of the address element, thereby being favorable for better solving the address service requirement based on the address element and improving the use experience of users aiming at the address element related functions and applications.
Example two
Referring to fig. 2, fig. 2 is a flow chart of another address element parsing method based on semantic expression according to an embodiment of the present invention. The method described in fig. 2 may be applied to an address element parsing apparatus based on semantic expression, where the apparatus may include a server, where the server includes a local server or a cloud server, and embodiments of the present invention are not limited. As shown in fig. 2, the address element parsing method based on semantic expression includes the following operations:
201. and acquiring training set data for address element analysis training, and constructing and training an address element analysis model based on the training set data to obtain the trained address element analysis model.
Alternatively, the address element parsing model may be constructed by one or more of a BERT model, a biglu model, and a CRF model, which is not limited by the embodiment of the present invention.
202. And when detecting that the address data needs to be subjected to element analysis operation, determining the standard address data corresponding to the address data.
203. And inputting the standard address data into the trained address element analysis model to obtain a corresponding address element analysis result.
Alternatively, the training process of the address element analysis model, the application process of the address element analysis model, and the connection relationship between the training process and the application process may be as shown in fig. 6.
It should be noted that, step 201 may be performed before the existence of address data is detected to require the element analysis operation, or may be performed after the existence of address data is detected to require the element analysis operation, which is not limited in the embodiment of the present invention.
In the embodiment of the present invention, for other descriptions of step 202 to step 203, please refer to other detailed descriptions of step 101 to step 102 in the first embodiment, and the description of the embodiment of the present invention is omitted.
Therefore, the embodiment of the invention can adopt the trained address element analysis model to realize element analysis of the address data, improve the running stability and reliability of the trained address element analysis model, be favorable for improving the determination accuracy and efficiency of the address element analysis result, and further improve the analysis accuracy and efficiency of the address element, thereby being favorable for better solving the address service requirement based on the address element and improving the use experience of users aiming at the address element related functions and applications; and the intelligent function of the address element analysis mode based on semantic expression is enriched, the comprehensiveness and the integrity of the address element analysis mode based on semantic expression are improved, the rationality and the feasibility of the address element analysis mode based on semantic expression are improved, the address element analysis model can be trained based on the scheme and corresponding training data, the fitting degree of the address element analysis model obtained through training and the scheme is improved, the running stability and the reliability of the address element analysis model are improved, and the analysis efficiency, the convenience and the accuracy of the address element are improved.
In an optional embodiment, the building and training an address element analysis model based on the training set data to obtain a trained address element analysis model may include:
constructing a basic analysis model according to the semantic analysis module, the hidden information acquisition module and the element constraint module;
training a basic analytical model according to the training set data to obtain a trained analytical model;
according to the address prediction result of address prediction of the training set data and the labeled address information corresponding to the training set data by the basic analysis model, calculating a model training loss value;
judging whether the model training loss value is smaller than or equal to a set model training loss value threshold value;
when the judgment result is yes, determining the trained analytic model as a trained address element analytic model;
and when the judgment result is negative, adjusting model training parameters, and executing model training operation based on the adjusted model training parameters until the model training loss value is smaller than or equal to a model training loss value threshold and/or the training times reach a preset times threshold.
Optionally, the semantic analysis module may be a module based on a BERT model, or may be based on another model, which is not limited in the embodiment of the present invention; the hidden information acquisition module can be a module based on a BIGRU network, and can also be other network models, and the embodiment of the invention is not limited; the element constraint module may be a module combining with the CRF method, or may be another method, which is not limited by the embodiment of the present invention.
Optionally, when the training frequency reaches the preset frequency threshold, the analysis model corresponding to the training frequency reaching the preset frequency threshold can be determined to be the address element analysis model after training, or the analysis model training operation can be ended, the historical training data of the analysis model is combined for feedback analysis and then training is performed again, so that the situation that the address element analysis model is obtained through malicious training can be reduced to a certain extent, and the reliability of the address element analysis model after training is improved.
Optionally, the labeling address information corresponding to the training set data may refer to a correct labeling result for model training, which is manually labeled and corresponds to the training set data.
Therefore, the optional embodiment can construct and train the analytic model, determine whether the analytic model meets the training completion condition, if yes, determine that the training of the address element analytic model is completed, if not, continuously adjust the training analytic model until convergence, thereby being beneficial to improving the comprehensiveness and rationality of the training mode of the analytic model, further improving the accuracy and reliability of the determined training result of the analytic model, further improving the reliability and running stability of the trained address element analytic model, and further improving the determination timeliness, efficiency and convenience of the training result of the analytic model, further improving the training efficiency of the address element analytic model and reducing unnecessary training waste.
In another optional embodiment, the training the basic analytical model according to the training set data to obtain a trained analytical model may include:
for any data to be trained in the training set data, inputting the data to be trained into a corresponding semantic analysis module to obtain a word vector representation result corresponding to characters in the data to be trained;
inputting the word vector representation result into a corresponding hidden information acquisition module to obtain a hidden representation result corresponding to the character, wherein the hidden representation result comprises a forward hidden semantic result and/or a reverse hidden semantic result, and determining front and rear semantic information corresponding to the character according to the hidden representation result and data obtained by performing data preprocessing operation on data to be trained;
and determining a corresponding address prediction result according to the element constraint module, the front and rear semantic information and the data to be trained, and taking a basic analysis model after outputting the address prediction result as a trained analysis model.
Optionally, the word vector representation result corresponding to the character can represent text semantic information corresponding to the character, which is not limited in the embodiment of the present invention.
For example, the word vector of each Chinese character is obtained and is input into a BIGRU network, and the BIGRU network models the hidden information of the current character information and the previous character through a gate control cyclic neural network to obtain the hidden representation of the current word; adopting two gating cyclic neural networks to respectively accept a forward input text and a reverse input text, and finally splicing forward and reverse hidden representations of a single Chinese character; the BIGRU network can effectively learn the front and back semantic information of the input address text, so that the model has the capability of learning the context semantics of the text.
Therefore, the optional embodiment can provide a mode of training the analytic model by specifically combining the data to be trained, which is beneficial to improving the comprehensiveness and rationality of the training mode of the analytic model, and further beneficial to improving the training efficiency, convenience, step order and feasibility of the analytic model, thereby being beneficial to improving the determining efficiency and reliability of the address element analytic model after the training is completed.
In still another optional embodiment, determining the corresponding address prediction result according to the element constraint module, the front-back semantic information and the data to be trained may include:
inputting the front and rear semantic information into an element constraint module to obtain at least one sequence tag prediction result corresponding to characters in data to be trained, and calculating to obtain a first validity corresponding to each sequence tag prediction result by combining a tag analysis algorithm;
determining a target sequence tag prediction result from all sequence tag prediction results according to the first validity and the set tag analysis conditions;
performing normalization processing operation on the first validity of each prediction label in the target sequence label prediction result, determining the second validity of each prediction label corresponding to the character, and determining the target prediction label of the character according to the second validity of all the prediction labels;
And determining an address prediction result corresponding to the data to be trained according to target prediction labels of all characters corresponding to the data to be trained.
For example, the context representation of each chinese character is obtained through biglu, to further learn constraint information of text, for example, to prevent the predicted text label from starting with "B-" and ending with "E-" but also to prevent "B-label", "I-label", "E-label" corresponding to the predicted address data from being different in "label" type; performing text constraint by adopting a CRF method, wherein the CRF method receives the context representation of each Chinese character output by the BIGRU, and calculates a sequence tag with the highest score by using a Viterbi algorithm; the essence of the address element analysis task is a classification task, the score of each label is output through CRF, and after normalization is carried out through a softmax activation function, the prediction probability of each label is obtained by each Chinese character, wherein the label with the maximum probability value is the prediction label of the current Chinese character.
Therefore, the optional embodiment can combine the element constraint module, the label analysis operation and the normalization operation to determine the validity of the sequence label prediction result and the validity of the prediction label, and further determine the address prediction result corresponding to the data to be trained, which is beneficial to improving the comprehensiveness and rationality of the determination mode of the address prediction result, further improving the determination efficiency and convenience of the address prediction result, and improving the accuracy and reliability of the determined address prediction result, thereby being beneficial to improving the determination efficiency and accuracy of the subsequent analysis model training result determined based on the address prediction result.
In still another optional embodiment, the determining the front and rear semantic information corresponding to the character according to the hidden representation result and the data obtained by performing the data preprocessing operation on the data to be trained may include:
when the hidden representation result comprises a forward hidden semantic result and a reverse hidden semantic result, determining corresponding forward input text and reverse input text according to data obtained by performing data preprocessing operation on data to be trained;
respectively inputting the forward input text and the reverse input text into a corresponding gating cyclic neural network to obtain a processing result;
and determining front and rear semantic information corresponding to the character according to the processing result, the forward latent semantic result, the reverse latent semantic result and the set latent semantic processing conditions.
Further optionally, determining the front and rear semantic information corresponding to the character according to the hidden representation result and the data obtained by performing the data preprocessing operation on the data to be trained may further include:
when the hidden representation result comprises a forward hidden semantic result or a reverse hidden semantic result, determining corresponding forward input text and reverse input text according to data obtained by performing data preprocessing operation on data to be trained; respectively inputting the forward input text and the reverse input text into a corresponding gating cyclic neural network to obtain a processing result;
And determining front and rear semantic information corresponding to the character according to the hidden representation result, the processing result, the set forward and reverse deduction analysis algorithm and the set hidden semantic processing condition.
Therefore, the optional embodiment can provide a matched character front and rear semantic information determining mode aiming at the condition that the hidden representation result comprises a forward result and a reverse result, which is beneficial to improving the comprehensiveness and rationality of the front and rear semantic information determining mode, further beneficial to improving the determining efficiency, convenience and accuracy of the front and rear semantic information, and further beneficial to improving the determining efficiency and accuracy of the address prediction result which is determined based on the front and rear semantic information; and the matched character front and rear semantic information determining mode can be provided according to the condition that the hidden representation result comprises the forward result or the reverse result, so that the diversity and flexibility of the character front and rear semantic information determining mode are enriched, and further the front and rear semantic information determining efficiency, convenience and accuracy are improved.
In still another optional embodiment, the calculating the model training loss value according to the address prediction result of performing address prediction on the training set data according to the base parsing model and the labeled address information corresponding to the training set data may include:
Determining an analysis validity result corresponding to each sub-parameter in the analysis judgment parameters according to an address prediction result of address prediction of the training set data, labeling address information corresponding to the training set data and the set analysis judgment parameters by the basic analysis model; the analysis judgment parameters comprise one or more of accuracy parameters, recall parameters, stability parameters, analysis efficiency parameters and analysis cost performance parameters;
according to the set parameter weighting conditions and analysis effectiveness results corresponding to all the sub-parameters, determining analysis effectiveness conditions corresponding to the basic analysis model, and determining corresponding model training loss values according to the analysis effectiveness conditions.
Optionally, the analysis and judgment parameters may further include a relationship parameter corresponding to at least two sub-parameters, for example, the relationship parameter may be an F1-score parameter, that is, a relationship parameter corresponding to the recall parameter and the accuracy parameter, which is not illustrated herein.
Therefore, the optional embodiment can provide a specific model training loss value determining mode, and determine the analysis validity condition of the analysis model by combining the analysis judgment parameters and the parameter weighting conditions, so as to determine the model training loss value, thereby being beneficial to improving the rationality and the comprehensiveness of the model training loss value determining mode, further being beneficial to improving the determining efficiency, the convenience and the accuracy of the model training loss value, and further being beneficial to improving the determining efficiency and the accuracy of the analysis model training result determined based on the model training loss value.
In yet another alternative embodiment, the specific manner of performing the data preprocessing operation on the data to be trained is as follows:
determining first processing data according to the data to be trained and the data cleaning conditions;
determining second processing data as data of which the data preprocessing operation is completed corresponding to the data to be trained according to the first processing data and the data format preprocessing conditions;
the data cleaning conditions comprise one or more of full angle and half angle conversion, unified bracket form, punctuation mark deletion, space deletion, foreign language letter case conversion, chinese character simplified conversion, unified language form and unified character form.
Optionally, the specific manner of performing the data preprocessing operation on the address data corresponding to the step of determining the standard address data corresponding to the address data may refer to the operation corresponding to the data cleaning condition and the data format preprocessing condition, or may also use other data preprocessing operations, which is not limited by the embodiment of the present invention.
Optionally, the data format corresponding to the data format preprocessing condition may be the above-mentioned BIES format using a word as a unit, or may be other text character formats, which is not limited in the embodiment of the present invention.
Therefore, the optional embodiment can provide a data preprocessing mode, is beneficial to improving the comprehensiveness, rationality and feasibility of data preprocessing, and further is beneficial to improving the data preprocessing efficiency and convenience, and in addition, improves the data normalization of the data to be trained, and is beneficial to improving the training efficiency and convenience of an analytical model based on the data after the data preprocessing to a certain extent.
In yet another alternative embodiment, the adjusting the model training parameters may include:
screening target subparameters from all subparameters according to all analysis validity results, and determining target data associated with the target subparameters from the address prediction results and the labeling address information according to the target subparameters;
and determining model training parameters to be adjusted according to the parameter adjustment algorithm, the target data and the analysis effectiveness results corresponding to the target subparameters, and adjusting the model training parameters.
Further optionally, adjusting the corresponding mode of the model training parameters may further include: and adjusting model training parameters according to the address prediction result corresponding to the address data, the real labeling result corresponding to the address data and the set back propagation algorithm to obtain an analysis model with optimal training effect.
Further optionally, the trained address element analysis model may also correspond to an analysis service and prediction module, where the analysis service loads an optimal model. Specifically, the overall address element analysis service adopts flash development, provides service to the outside in an http request mode, and simultaneously adopts a multi-process and asynchronous mode during service development to improve the request performance of the service.
In the above optional embodiment, further optionally, the screening the target sub-parameters from all sub-parameters according to all analysis validity results may include:
when the analysis validity result is used for representing the analysis validity corresponding to the target subparameter, the subparameter with the analysis validity smaller than or equal to a preset analysis validity threshold value is screened out from all the subparameters, and/or the sorting condition of the analysis validity corresponding to all the subparameters is determined, and the subparameter meeting the sorting condition is determined according to the sorting condition to serve as the target subparameter.
Therefore, the optional embodiment can provide a specific model training parameter adjustment mode, is beneficial to improving the rationality and the comprehensiveness of the model training parameter adjustment mode, is beneficial to improving the adjustment accuracy and the reliability of the model training parameter, and improves the adjustment efficiency and the convenience of the model training parameter, thereby being beneficial to improving the completion efficiency of the analytical model training; and the target sub-parameters can be determined through the threshold comparison conditions and/or the conditions before and after sequencing, so that the diversity and flexibility of the target sub-parameter determination mode are improved, the determination efficiency, convenience and accuracy of the target sub-parameters are improved, and the efficiency and accuracy of the subsequent model training parameter adjustment based on the target sub-parameters are improved.
Example III
Referring to fig. 3, fig. 3 is a schematic structural diagram of an address element parsing device based on semantic expression according to an embodiment of the present invention. The apparatus described in fig. 3 may include a server, where the server includes a local server or a cloud server, and embodiments of the present invention are not limited. As shown in fig. 3, the address element parsing apparatus based on semantic expression may include:
a determining module 301, configured to determine canonical address data corresponding to the address data when it is detected that the address data needs to perform element parsing operation; the standard address data is address data expressed in a set text format after data preprocessing.
The element analysis module 302 is configured to input the canonical address data into the trained address element analysis model to obtain a corresponding address element analysis result; the address element analysis result is an analysis result obtained after the character division operation is performed on the basis of the determined predicted address element category corresponding to the standard address data.
Therefore, the address element analysis device based on semantic expression described in fig. 3 can adopt the trained address element analysis model to realize element analysis of the address data, so that the running stability and reliability of the trained address element analysis model are improved, the determination accuracy and efficiency of the address element analysis result are improved, the analysis accuracy and efficiency of the address element are improved, and the address service requirement based on the address element is better solved, so that the use experience of users for functions and applications related to the address element is improved.
In an alternative embodiment, as shown in fig. 4, the apparatus may further include:
the information obtaining module 303 is configured to obtain training set data for address element analysis training before the element analysis module 302 inputs the canonical address data into the trained address element analysis model to obtain a corresponding address element analysis result.
The model training module 304 is configured to construct and train an address element analysis model based on the training set data, and obtain a trained address element analysis model.
Therefore, the device described in fig. 4 can provide the construction and training modes of the address element analysis model, enrich the intelligent functions of the address element analysis mode based on semantic expression, and facilitate the improvement of the comprehensiveness and the integrity of the address element analysis mode based on semantic expression, further facilitate the improvement of the rationality and the feasibility of the address element analysis mode based on semantic expression, and further facilitate the training of the address element analysis model based on the scheme and the corresponding training data, improve the fitting degree of the address element analysis model obtained by training and the scheme, and facilitate the improvement of the operation stability and the reliability of the address element analysis model, and further facilitate the improvement of the analysis efficiency, the convenience and the accuracy of the address element.
In another alternative embodiment, the model training module 304 builds and trains the address element analysis model based on the training set data, and the method for obtaining the trained address element analysis model specifically includes:
constructing a basic analysis model according to the semantic analysis module, the hidden information acquisition module and the element constraint module;
training a basic analytical model according to the training set data to obtain a trained analytical model;
according to the address prediction result of address prediction of the training set data and the labeled address information corresponding to the training set data by the basic analysis model, calculating a model training loss value;
judging whether the model training loss value is smaller than or equal to a set model training loss value threshold value;
when the judgment result is yes, determining the trained analytic model as a trained address element analytic model;
and when the judgment result is negative, adjusting model training parameters, and executing model training operation based on the adjusted model training parameters until the model training loss value is smaller than or equal to a model training loss value threshold and/or the training times reach a preset times threshold.
It can be seen that the device described in fig. 4 can also be used for constructing and training the analytical model, determining that the training of the address element analytical model is completed if the analytical model meets the training completion condition, and if not, continuing to adjust the training analytical model until convergence, thereby being beneficial to improving the comprehensiveness and rationality of the training mode of the analytical model, further improving the accuracy and reliability of the determined training result of the analytical model, further improving the reliability and operation stability of the trained address element analytical model, and further improving the timeliness, efficiency and convenience of the determination of the training result of the analytical model, further improving the training efficiency of the address element analytical model and reducing unnecessary training waste.
In yet another alternative embodiment, the model training module 304 trains the basic analytical model according to the training set data, and the method for obtaining the trained analytical model specifically includes:
for any data to be trained in the training set data, inputting the data to be trained into a corresponding semantic analysis module to obtain a word vector representation result corresponding to characters in the data to be trained;
inputting the word vector representation result into a corresponding hidden information acquisition module to obtain a hidden representation result corresponding to the character, wherein the hidden representation result comprises a forward hidden semantic result and/or a reverse hidden semantic result, and determining front and rear semantic information corresponding to the character according to the hidden representation result and data obtained by performing data preprocessing operation on data to be trained;
and determining a corresponding address prediction result according to the element constraint module, the front and rear semantic information and the data to be trained, and taking a basic analysis model after outputting the address prediction result as a trained analysis model.
Therefore, the device described in fig. 4 can also provide a mode of training the analytical model by specifically combining the data to be trained, which is favorable for improving the comprehensiveness and rationality of the training mode of the analytical model, and further is favorable for improving the training efficiency, convenience, step order and feasibility of the analytical model, thereby being favorable for improving the determining efficiency and reliability of the address element analytical model after the training is completed.
In yet another optional embodiment, the mode of determining the corresponding address prediction result by the model training module 304 according to the element constraint module, the front-back semantic information and the data to be trained specifically includes:
inputting the front and rear semantic information into an element constraint module to obtain at least one sequence tag prediction result corresponding to characters in data to be trained, and calculating to obtain a first validity corresponding to each sequence tag prediction result by combining a tag analysis algorithm;
determining a target sequence tag prediction result from all sequence tag prediction results according to the first validity and the set tag analysis conditions;
performing normalization processing operation on the first validity of each prediction label in the target sequence label prediction result, determining the second validity of each prediction label corresponding to the character, and determining the target prediction label of the character according to the second validity of all the prediction labels;
and determining an address prediction result corresponding to the data to be trained according to target prediction labels of all characters corresponding to the data to be trained.
It can be seen that the device described in fig. 4 can also be implemented to combine the element constraint module, the tag analysis operation and the normalization operation to determine the validity of the sequence tag prediction result and the validity of the prediction tag, and further determine the address prediction result corresponding to the data to be trained, which is beneficial to improving the comprehensiveness and rationality of the determination mode of the address prediction result, further improving the determination efficiency and convenience of the address prediction result, and improving the accuracy and reliability of the determined address prediction result, thereby improving the determination efficiency and accuracy of the subsequent analysis model training result determined based on the address prediction result.
In yet another optional embodiment, the mode of determining the front and rear semantic information corresponding to the character by the model training module 304 according to the hidden representation result and the data obtained by performing the data preprocessing operation on the data to be trained specifically includes:
when the hidden representation result comprises a forward hidden semantic result and a reverse hidden semantic result, determining corresponding forward input text and reverse input text according to data obtained by performing data preprocessing operation on data to be trained;
respectively inputting the forward input text and the reverse input text into a corresponding gating cyclic neural network to obtain a processing result;
and determining front and rear semantic information corresponding to the character according to the processing result, the forward latent semantic result, the reverse latent semantic result and the set latent semantic processing conditions.
Therefore, the device described in fig. 4 can also provide a matched character front-rear semantic information determining mode according to the situation that the hidden representation result comprises the forward result and the reverse result, which is beneficial to improving the comprehensiveness and rationality of the front-rear semantic information determining mode, further beneficial to improving the determining efficiency, convenience and accuracy of the front-rear semantic information, and further beneficial to improving the determining efficiency and accuracy of the address prediction result determined based on the front-rear semantic information.
In yet another optional embodiment, the mode of calculating the model training loss value by the model training module 304 according to the address prediction result of the address prediction of the basic analytical model on the training set data and the labeled address information corresponding to the training set data specifically includes:
determining an analysis validity result corresponding to each sub-parameter in the analysis judgment parameters according to an address prediction result of address prediction of the training set data, labeling address information corresponding to the training set data and the set analysis judgment parameters by the basic analysis model; the analysis judgment parameters comprise one or more of accuracy parameters, recall parameters, stability parameters, analysis efficiency parameters and analysis cost performance parameters;
according to the set parameter weighting conditions and analysis effectiveness results corresponding to all the sub-parameters, determining analysis effectiveness conditions corresponding to the basic analysis model, and determining corresponding model training loss values according to the analysis effectiveness conditions.
Therefore, the device described in fig. 4 can also provide a specific model training loss value determining manner, and determine the analysis validity of the analysis model in combination with the analysis judgment parameters and the parameter weighting conditions, so as to determine the model training loss value, which is beneficial to improving the rationality and the comprehensiveness of the model training loss value determining manner, and further beneficial to improving the determining efficiency, the convenience and the accuracy of the model training loss value, so as to be beneficial to improving the determining efficiency and the accuracy of the analysis model training result determined based on the model training loss value.
In yet another alternative embodiment, the specific way of performing the data preprocessing operation on the data to be trained is:
determining first processing data according to the data to be trained and the data cleaning conditions;
determining second processing data as data of which the data preprocessing operation is completed corresponding to the data to be trained according to the first processing data and the data format preprocessing conditions;
the data cleaning conditions comprise one or more of full angle and half angle conversion, unified bracket form, punctuation mark deletion, space deletion, foreign language letter case conversion, chinese character simplified conversion, unified language form and unified character form.
Therefore, the device described in fig. 4 can also provide a data preprocessing mode, which is favorable for improving the comprehensiveness, rationality and feasibility of data preprocessing, further is favorable for improving the data preprocessing efficiency and convenience, and further, improves the data normalization of the data to be trained, thereby being favorable for improving the training efficiency and convenience of the analytic model based on the data after the data preprocessing to a certain extent.
Example IV
Referring to fig. 5, fig. 5 is a schematic structural diagram of another address element parsing device based on semantic expression according to an embodiment of the present invention. The apparatus described in fig. 5 may include a server, where the server includes a local server or a cloud server, and embodiments of the present invention are not limited. As shown in fig. 5, the apparatus may include:
A memory 401 storing executable program codes;
a processor 402 coupled with the memory 401;
further, an input interface 403 and an output interface 404 coupled to the processor 402 may be included;
wherein the processor 402 invokes executable program code stored in the memory 401 for performing the steps in the semantic expression based address element parsing method described in embodiment one or embodiment two.
Example five
The embodiment of the invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program causes a computer to execute the steps in the address element analysis method based on semantic expression described in the first embodiment or the second embodiment.
Example six
An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps in the address element parsing method based on semantic expression described in the first or second embodiment.
The apparatus embodiments described above are merely illustrative, wherein the modules illustrated as separate components may or may not be physically separate, and the components shown as modules may or may not be physical, i.e., may be located in one place, or may be distributed over a plurality of network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above detailed description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course by means of hardware. Based on such understanding, the foregoing technical solutions may be embodied essentially or in part in the form of a software product that may be stored in a computer-readable storage medium including Read-Only Memory (ROM), random-access Memory (Random Access Memory, RAM), programmable Read-Only Memory (Programmable Read-Only Memory, PROM), erasable programmable Read-Only Memory (Erasable Programmable Read Only Memory, EPROM), one-time programmable Read-Only Memory (OTPROM), electrically erasable programmable Read-Only Memory (EEPROM), compact disc Read-Only Memory (Compact Disc Read-Only Memory, CD-ROM) or other optical disc Memory, magnetic disc Memory, tape Memory, or any other medium that can be used for computer-readable carrying or storing data.
Finally, it should be noted that: the embodiment of the invention discloses an address element analysis method and device based on semantic expression, which are disclosed by the embodiment of the invention and are only used for illustrating the technical scheme of the invention, but not limiting the technical scheme; although the invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art will understand that; the technical scheme recorded in the various embodiments can be modified or part of technical features in the technical scheme can be replaced equivalently; such modifications and substitutions do not depart from the spirit and scope of the corresponding technical solutions.

Claims (10)

1. An address element parsing method based on semantic expression, which is characterized by comprising the following steps:
when detecting that address data need to be subjected to element analysis operation, determining standard address data corresponding to the address data; the standard address data are address data expressed in a set text format after data preprocessing;
inputting the standard address data into a trained address element analysis model to obtain a corresponding address element analysis result; and the address element analysis result is an analysis result which is obtained after the character division operation is performed on the basis of the determined predicted address element category and corresponds to the standard address data.
2. The semantic expression-based address element parsing method according to claim 1, wherein before the inputting the canonical address data into the trained address element parsing model, the method further comprises:
acquiring training set data for address element analysis training, and constructing and training an address element analysis model based on the training set data to obtain the address element analysis model after training;
and constructing and training an address element analysis model based on the training set data to obtain the trained address element analysis model, wherein the method comprises the following steps:
constructing a basic analysis model according to the semantic analysis module, the hidden information acquisition module and the element constraint module;
training the basic analytic model according to the training set data to obtain a trained analytic model;
according to the address prediction result of the address prediction of the training set data by the basic analysis model and the labeled address information corresponding to the training set data, calculating a model training loss value;
judging whether the model training loss value is smaller than or equal to a set model training loss value threshold value or not;
When the judgment result is yes, determining the trained analytic model as a trained address element analytic model;
and when the judgment result is negative, adjusting model training parameters, and executing model training operation based on the adjusted model training parameters until the model training loss value is smaller than or equal to the model training loss value threshold and/or the training times reach a preset time threshold.
3. The semantic expression-based address element parsing method according to claim 2, wherein training the basic parsing model according to the training set data to obtain a trained parsing model comprises:
for any data to be trained in the training set data, inputting the data to be trained into the corresponding semantic analysis module to obtain a word vector representation result corresponding to characters in the data to be trained;
inputting the word vector representation result into the corresponding hidden information acquisition module to obtain a hidden representation result corresponding to the character, wherein the hidden representation result comprises a forward hidden semantic result and/or a reverse hidden semantic result, and determining front and rear semantic information corresponding to the character according to the hidden representation result and data obtained by performing data preprocessing operation on the data to be trained;
And determining a corresponding address prediction result according to the element constraint module, the front and rear semantic information and the data to be trained, and taking a basic analysis model after outputting the address prediction result as a trained analysis model.
4. The semantic expression-based address element parsing method according to claim 3, wherein determining a corresponding address prediction result according to the element constraint module, the front-back semantic information and the data to be trained includes:
inputting the front and rear semantic information into the element constraint module to obtain at least one sequence tag prediction result corresponding to the characters in the data to be trained, and calculating to obtain a first validity corresponding to each sequence tag prediction result by combining a tag analysis algorithm;
determining a target sequence tag prediction result from all the sequence tag prediction results according to the first validity and the set tag analysis conditions;
performing normalization processing operation on the first validity corresponding to each prediction tag in the target sequence tag prediction result, determining the second validity of each prediction tag corresponding to the character, and determining the target prediction tag of the character according to the second validity of all the prediction tags;
And determining an address prediction result corresponding to the data to be trained according to target prediction labels of all the characters corresponding to the data to be trained.
5. The method for resolving an address element based on semantic expression according to claim 3, wherein determining the front and rear semantic information corresponding to the character according to the hidden representation result and the data obtained by performing the data preprocessing operation on the data to be trained comprises:
when the hidden representation result comprises the forward hidden semantic result and the reverse hidden semantic result, determining corresponding forward input text and reverse input text according to data obtained by performing data preprocessing operation on the data to be trained;
respectively inputting the forward input text and the reverse input text into corresponding gating cyclic neural networks to obtain a processing result;
and determining front and rear semantic information corresponding to the character according to the processing result, the forward latent semantic result, the reverse latent semantic result and the set latent semantic processing conditions.
6. The semantic expression-based address element parsing method according to claim 4 or 5, wherein the calculating a model training loss value according to the address prediction result of the address prediction of the training set data by the base parsing model and the labeled address information corresponding to the training set data includes:
Determining an analysis validity result corresponding to each sub-parameter in the analysis judgment parameters according to an address prediction result of address prediction of the training set data by the basic analysis model, labeling address information corresponding to the training set data and the set analysis judgment parameters; the analysis judgment parameters comprise one or more of accuracy parameters, recall rate parameters, stability parameters, analysis efficiency parameters and analysis cost performance parameters;
and determining the analysis validity conditions corresponding to the basic analysis model according to the set parameter weighting conditions and the analysis validity results corresponding to all the sub-parameters, and determining the corresponding model training loss values according to the analysis validity conditions.
7. The semantic expression-based address element parsing method according to claim 6, wherein the specific way of performing the data preprocessing operation on the data to be trained is:
determining first processing data according to the data to be trained and the data cleaning conditions;
determining second processing data as data of which the data preprocessing operation is completed corresponding to the data to be trained according to the first processing data and the data format preprocessing conditions;
The data cleaning conditions comprise one or more of full-angle and half-angle conversion, unified bracket form, punctuation mark deletion, space deletion, foreign language letter case conversion, chinese character simplified conversion, unified language form and unified character form.
8. An address element parsing apparatus based on semantic expression, the apparatus comprising:
the determining module is used for determining the standard address data corresponding to the address data when detecting that the address data need to be subjected to element analysis operation; the standard address data are address data expressed in a set text format after data preprocessing;
the element analysis module is used for inputting the standard address data into the trained address element analysis model to obtain a corresponding address element analysis result; and the address element analysis result is an analysis result which is obtained after the character division operation is performed on the basis of the determined predicted address element category and corresponds to the standard address data.
9. An address element parsing apparatus based on semantic expression, the apparatus comprising:
a memory storing executable program code;
a processor coupled to the memory;
The processor invokes the executable program code stored in the memory to perform the semantic expression based address element resolution method of any one of claims 1-7.
10. A computer storage medium storing computer instructions which, when invoked, are operable to perform the semantic expression based address element resolution method of any one of claims 1-7.
CN202211557672.3A 2022-12-06 2022-12-06 Address element analysis method and device based on semantic expression Pending CN116227460A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211557672.3A CN116227460A (en) 2022-12-06 2022-12-06 Address element analysis method and device based on semantic expression

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211557672.3A CN116227460A (en) 2022-12-06 2022-12-06 Address element analysis method and device based on semantic expression

Publications (1)

Publication Number Publication Date
CN116227460A true CN116227460A (en) 2023-06-06

Family

ID=86577446

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211557672.3A Pending CN116227460A (en) 2022-12-06 2022-12-06 Address element analysis method and device based on semantic expression

Country Status (1)

Country Link
CN (1) CN116227460A (en)

Similar Documents

Publication Publication Date Title
CN111090727B (en) Language conversion processing method and device and dialect voice interaction system
CN112100349A (en) Multi-turn dialogue method and device, electronic equipment and storage medium
CN111402861B (en) Voice recognition method, device, equipment and storage medium
CN108710704B (en) Method and device for determining conversation state, electronic equipment and storage medium
JP6677419B2 (en) Voice interaction method and apparatus
CN110795532A (en) Voice information processing method and device, intelligent terminal and storage medium
US11749255B2 (en) Voice question and answer method and device, computer readable storage medium and electronic device
WO2022095354A1 (en) Bert-based text classification method and apparatus, computer device, and storage medium
WO2022048194A1 (en) Method, apparatus and device for optimizing event subject identification model, and readable storage medium
CN113076720B (en) Long text segmentation method and device, storage medium and electronic device
CN115840808B (en) Technological project consultation method, device, server and computer readable storage medium
CN116991875B (en) SQL sentence generation and alias mapping method and device based on big model
CN116304717B (en) Text classification method and device, storage medium and electronic equipment
CN112036186A (en) Corpus labeling method and device, computer storage medium and electronic equipment
CN113705207A (en) Grammar error recognition method and device
CN116306663B (en) Semantic role labeling method, device, equipment and medium
CN115859999B (en) Intention recognition method, device, electronic equipment and storage medium
CN116680385A (en) Dialogue question-answering method and device based on artificial intelligence, computer equipment and medium
CN111128122A (en) Method and system for optimizing rhythm prediction model
CN116978367A (en) Speech recognition method, device, electronic equipment and storage medium
CN116227460A (en) Address element analysis method and device based on semantic expression
CN115691503A (en) Voice recognition method and device, electronic equipment and storage medium
CN115527520A (en) Anomaly detection method, device, electronic equipment and computer readable storage medium
CN114781359A (en) Text error correction method and device, computer equipment and storage medium
CN113850383A (en) Text matching model training method and device, electronic equipment and storage medium

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