Sustainably improvable corpus problem classification method and system

Abstract

The application provides a sustainable improvement corpus classification method, and belongs to the technical field of information processing and corpus construction. The method is based on cluster implementation and comprises the following steps: judging whether parallel corpus is generated by an electronic device; adopting a verification electronic device to verify a device set; performing pre-submission operation on each piece of parallel corpus saved in a pre-submitted corpus; updating a standard corpus according to a feedback result of the pre-submission operation, wherein the updating comprises the following steps: reclassifying parallel corpus of the standard corpus according to source corpus or target corpus. The cluster is a translation server cluster or a question and answer server cluster. The technical scheme of the application can continuously improve and update the constructed corpus information, and the updating process can effectively avoid corpus input errors caused by subjective deviation accumulation of individual users or devices. In addition, the updating process is based on a distributed transaction processing mechanism, thereby ensuring the consistency of the standard.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of information processing and corpus technology, and particularly relates to a corpus problem classification method and system that can be continuously improved, a computer readable storage medium for implementing the method, a computer program product, and an electronic device. BACKGROUND

[0002] Corpus refers to a large-scale electronic database that is scientifically sampled and processed, in which real language materials that actually appear in the use of language are stored. The initial corpus only provides a large number of natural and real language facts and text data for language research and teaching, for example, the corpus of parallel corpus can constitute a translation relationship and is mainly used in the application fields of machine translation and bilingual dictionary compilation. The data-driven learning method based on the corpus conforms to the learning theory of schema structure and constructivism, can help learners improve language awareness, internalize language use rules, and improve language communication ability, and the corpus has gradually become one of the important auxiliary tools for foreign language self-learning.

[0003] With the development of artificial intelligence technology, the application of corpus in other industries also gradually shows advantages. Researchers can extract and analyze the language features of various types of texts based on detailed corpus text evidence, more scientifically and objectively describe the lexical characteristics and chapter features of specific styles and domains, and combine a deep learning engine to construct a parallel corpus including questions and answers, realize personified computer-aided human-computer interaction, and the most common ones are intelligent customer service systems and intelligent robot dialogue systems.

[0004] No matter which application scenario, constructing a suitable corpus and accurately classifying problems is the premise, among which the accuracy of corpus entry, the accuracy and consistency of annotation in the corpus are important judgment standards. Chinese invention patent application CN201710020573.4 proposes a Chinese reference corpus construction method based on events, which claims to be established on the basis of Chinese syntactic analysis and semantic analysis, in line with the characteristics of Chinese, and the method can also check the consistency of the annotated corpus to ensure the quality of corpus annotation.

[0005] However, the evolution speed of natural language is relatively fast, and the corresponding corpus should also be updated in a timely manner. On the basis of the existing standard corpus, the corpus can be continuously improved through artificial auxiliary updating and machine automatic periodic updating. In this process, the first thing to ensure is the standard consistency of corpus updating; secondly, there may be differences in the standards for whether the same to-be-entered corpus is suitable for entry and updating by different users and different machines, and under the cluster effect, how to effectively avoid the accumulation of subjective bias of individual users or devices leading to corpus entry errors is also a technical problem to be solved. SUMMARY

[0006] In order to solve the above technical problems, the present application provides a sustainable improvement corpus question classification method and system, a computer readable storage medium for implementing the method, a computer program product and an electronic device.

[0007] In the first aspect of the present application, a sustainable improvement corpus question classification method is provided, which is based on a cluster implementation, and the cluster includes N electronic devices.

[0008] The method can be automatically executed based on a computer program, and the method includes the following computer executable steps:

[0009] S110: let i = 1;

[0010] S120: determine whether parallel corpus is generated by the electronic device; the parallel corpus includes source corpus and target corpus;

[0011] If yes, save the parallel corpus to a pre-submitted corpus, and go to step S130;

[0012] If no, add the electronic device to a verification device set as a verification electronic device, and go to step S130;

[0013] S130: determine whether i < N is true, if yes, let i = i + 1, and return to step S120;

[0014] If no, go to step S140;

[0015] S140: use the verification electronic device of the verification device set to perform a pre-submission operation on each piece of parallel corpus saved in the pre-submitted corpus;

[0016] S150: update a standard corpus according to a feedback result of the pre-submission operation, and the update includes reclassifying parallel corpus of the standard corpus according to source corpus or target corpus.

[0017] In one specific application, the cluster is a translation server cluster, the electronic device is configured with at least one translation engine; the parallel corpus is a translation corpus, the source corpus is a to-be-translated corpus, the target corpus is a translated corpus, and the to-be-translated corpus and the translated corpus are in different languages.

[0018] At this time, the step S140 specifically includes:

[0019] The to-be-translated corpus corresponding to each piece of translation corpus is used as the input of the translation engine of each verification electronic device, and the feedback result of the pre-submission operation is determined based on the comparison result of the output result of the translation engine and the translated corpus corresponding to each piece of translation corpus.

[0020] Different verification electronic devices call different translation engines each time.

[0021] In another specific application, the cluster is a question and answer server cluster; the parallel corpus is a question and answer corpus, the source corpus is a question corpus, and the target corpus is an answer corpus; the question corpus and the answer corpus are in the same language.

[0022] At this time, each electronic device is configured with an intelligent question and answer engine; different verification electronic devices call different intelligent question and answer engines each time.

[0023] The step S140 specifically includes:

[0024] The question corpus corresponding to each piece of question and answer corpus is taken as the input of the intelligent question and answer engine of each verification electronic device, and the feedback result of the pre-submission operation is determined based on the comparison result of the output result of the intelligent question and answer engine and the answer corpus corresponding to each piece of question and answer corpus.

[0025] In the second aspect of the present application, to realize the sustainable improvement of the corpus question classification method in the first aspect, a sustainable improvement of the corpus question classification system is provided, the system includes a cluster server and a standard corpus, the cluster server is in wireless communication with N electronic devices, and the system further includes:

[0026] A corpus detection unit is configured to detect whether the electronic device generates a parallel corpus;

[0027] A pre-submission unit is configured to save the parallel corpus generated by the electronic device to a pre-submission corpus;

[0028] A verification device designation unit is configured to designate the electronic device as a verification electronic device when the electronic device does not generate a parallel corpus;

[0029] A pre-operation execution unit is configured to perform a pre-submission operation on each piece of parallel corpus saved in the pre-submission corpus by the verification electronic device;

[0030] An updating unit is configured to update the standard corpus according to the feedback result of the pre-submission operation, and the updating includes reclassifying the parallel corpus of the standard corpus according to the source corpus or the target corpus;

[0031] The cluster server is a translation server and / or a question and answer server; the electronic device is configured with a translation engine or an intelligent question and answer engine; and N is an integer greater than 2.

[0032] The verification device designation unit constitutes a distributed transaction execution unit with all the verification electronic devices; the pre-commit operation of each parallel corpus saved in the pre-commit corpus library constitutes a distributed atomic transaction operation.

[0033] The aforementioned sustainable improvement corpus problem classification method can be realized automatically by various forms of electronic devices through computer program instructions; the computer program instructions can be stored in different forms of storage media and loaded into computer electronic devices for execution.

[0034] Therefore, in the third aspect of the present application, a computer readable storage medium is also provided for storing computer instructions, which, when executed on an electronic device, cause the electronic device to perform all or part of the steps of the aforementioned sustainable improvement corpus problem classification method.

[0035] In the fourth aspect of the present application, a computer device is also provided, which includes a processor and a memory, the memory being used to store instructions, and the processor being used to call the instructions in the memory, so that the computer device performs the aforementioned sustainable improvement corpus problem classification method.

[0036] In the fifth aspect of the present application, a computer program product is also provided, which includes a computer program, when the computer program is executed, all or part of the steps of the aforementioned sustainable improvement corpus problem classification method are realized.

[0037] The technical solution of the present application can continuously improve and update the constructed corpus information, and the updating process can effectively avoid corpus entry errors caused by the accumulation of subjective bias of individual users or devices; in addition, the updating process is based on a distributed transaction processing mechanism, which ensures the consistency of the standard.

[0038] The further advantages of the present application will be further embodied in detail in the specific embodiment part combined with the drawings of the specification. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below, and obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0040] Figure 1 is a sustainable improvement corpus problem classification method of an embodiment of the present application

[0041] Figure 2 is Figure 1 The method is particularly applied to the preferred embodiment of updating a question-answer corpus

[0042] Figure 3 is Figure 1 The method is particularly applied to the preferred embodiment of updating a question-answer corpus

[0043] Figure 4 is Figure 1 The hardware / software architecture diagram of one implementation scenario of the method

[0044] Figure 5 The functional module unit composition diagram of a sustainable improvement corpus question classification system according to one embodiment of the application DETAILED DESCRIPTION

[0045] First of all, it needs to be pointed out that the embodiments of the sustainable improvement corpus question classification method mentioned in this part can be realized on an electronic device, system configured with a memory and a processor through a computer program. The electronic device, system can be a physical machine, a virtual machine, a server, a cluster or any combination thereof.

[0046] Preferably, the specific form of the electronic device can also be a human-computer interaction terminal, which can be a desktop terminal, a smart handheld terminal, a mobile terminal, etc. with a human-computer interaction interface.

[0047] First of all, refer to Figure 1 , Figure 1 The main flow diagram of a sustainable improvement corpus question classification method according to one embodiment of the application is shown.

[0048] Figure 1 The method is realized based on a cluster, and the cluster includes N electronic devices, N>1. The computer implementation process of the method is as follows (for the convenience of understanding, all the figures omit the step number):

[0049] S110: Let i=1;

[0050] S120: Determine whether the electronic device generates parallel corpus; the parallel corpus includes source corpus and target corpus;

[0051] If yes, save the parallel corpus to the pre-submitted corpus, and go to step S130;

[0052] If no, add the electronic device as a verification electronic device to the verification device set, and go to step S130;

[0053] S130: Determine whether i<N is true, if yes, let i=i+1, and return to step S120;

[0054] If no, go to step S140;

[0055] S140: using the verification electronic device of the verification device set, pre-submits each parallel corpus saved by the pre-submission corpus;

[0056] S150: according to the feedback result of the pre-submission operation, updating the standard corpus, the updating includes reclassifying the parallel corpus of the standard corpus according to the source corpus or the target corpus.

[0057] Figure 1 The method generally summarizes the main process of the continuously improved corpus problem classification method of the application, which can be applied to all scenarios that need corpus support, including human-computer interaction translation, human-computer interaction dialogue, intelligent robot customer service, etc. In these scenarios, a standard corpus is pre-constructed, and based on the standard corpus, a corresponding artificial intelligence engine (translation engine, intelligent customer service engine, human-computer dialogue engine, etc.) is used to execute the corresponding human-computer interaction / dialogue process.

[0058] Due to the continuous updating of application scenarios and natural language, especially in recent years under the wave of big data and Internet+, the corpus changes day by day, and the original standard corpus also needs to be continuously updated and supplemented, otherwise it cannot keep up with the latest interaction trend, and the user experience will also be reduced.

[0059] This continuous supplement and update process is the continuous improvement process of the corpus problem and its classification. This process is usually maintained by a dedicated person or automatically collected and maintained by a machine. However, single dedicated maintenance is limited in quantity, so in many application scenarios, user participation or task crowdsourcing / sub-contracting mode is developed; the former encourages a large number of users in the corresponding scenario to actively report new words, new explanations, new translations, new answers / new questions, and then selects to enter the corpus; or also uses a crowdsourcing incentive mode, sub-contracts corpus update tasks to crowdsourcing users, and gives certain rewards to users who actively report corpus updates.

[0060] No matter which mode is used, since it involves multiple users / user terminal devices, there may be multiple different to-be-submitted results for the same corpus, and then the standard consistency problem of corpus update is generated. Because different users and different machines may have different judgment standards when facing the same to-be-entered corpus whether it is suitable for entering the library, under the cluster effect, how to effectively avoid the accumulation of subjective bias of individual users or devices leading to corpus entry errors is a technical problem that needs to be solved.

[0061] Figure 1Embodiments of the present application are proposed to solve the above problems. Among them, N electronic devices {Dev1, Dev2, …, Dev N} correspond to the aforementioned multiple users / user terminal devices, that is, crowd-sourced users / devices.

[0062] For example, a translation APP application is taken as an example. The translation APP realizes computer-aided translation based on the support of a cluster translation server. The cluster translation server is connected with a standard translation corpus. The standard translation corpus stores parallel corpus. The parallel corpus includes source corpus and target corpus. The source corpus and the target corpus have a corresponding correlation relationship. The corresponding relationship includes one-to-one, one-to-many, many-to-many or many-to-one. That is, one source corpus can correspond to one or more target corpora, and multiple source corpora can also correspond to one or more target corpora.

[0063] In this scenario, the parallel corpus is specifically a translation corpus, the source corpus is a to-be-translated corpus, the target corpus is a translated corpus, and the to-be-translated corpus and the translated corpus are different languages.

[0064] When a user inputs a to-be-translated word W1, a to-be-translated sentence S1 (to-be-translated corpus) of a first language, the translation APP calls the standard translation corpus through the cluster translator server to execute a translation engine and outputs a corresponding translation result as a translated word W2, a translated sentence S2 (target corpus) of a second language.

[0065] For example, a second APP application supporting intelligent man-machine customer service is taken as an example. The second APP can provide intelligent man-machine customer service. When a user has a related problem in using the APP and needs to consult, the second APP can call a question and answer server cluster based on a standard question and answer corpus to independently respond to the user's input question. Each parallel corpus in the parallel corpus included in the standard question and answer corpus is also composed of a source corpus and a target corpus. The parallel corpus is a question and answer corpus, the source corpus is a question corpus, and the target corpus is an answer corpus. The question corpus and the answer corpus are of the same language. The source corpus and the target corpus have a corresponding correlation relationship. The corresponding relationship can also include one-to-one, one-to-many, many-to-many or many-to-one. That is, one source corpus can correspond to one or more target corpora, and multiple source corpora can also correspond to one or more target corpora.

[0066] No matter whether it is a standard translation corpus or a standard question and answer corpus, the corpus and the question stored therein need to be updated in a timely manner, and the question (source corpus) or the answer (target corpus) also needs to be reclassified according to the situation, so as to follow the latest interactive trend and update the user experience.

[0067] Figure 1A sustainable improved corpus problem classification method is generally given, in which N users correspond to N electronic devices using a certain standard corpus; the N users are encouraged to actively submit new corpus to update the standard corpus. In the prior art, different users randomly submit corpus, and the submitted corpus has different values and different judgment standards.

[0068] To change this situation, Figure 1 The method is improved in at least three aspects:

[0069] (1) The user (device / terminal) who submits the corpus is separated from the user (device / terminal) who verifies the corpus: when a user (device / terminal) submits a corpus, the user (device / terminal) cannot serve as a verification device to achieve cross-validation, step S120 corresponds to the corresponding technical means.

[0070] (2) The corpus needs to be pre-submitted before formal submission to ensure standard consistency; step S140 corresponds to the corresponding technical means.

[0071] (3) According to the feedback result of the pre-submission operation, update the standard corpus, embody the distributed atomic transaction operation, step S150 corresponds to the corresponding technical means.

[0072] As a further preferred embodiment, to improve the execution efficiency of the method, the N electronic devices are the current active users of the current standard corpus corresponding application, that is, the N electronic devices are currently using the application (the aforementioned translation APP or the second application APP), so as to ensure the active process of data interaction.

[0073] Figure 1 The embodiments of the application are embodiments of the improved concept of the application as a whole. Next, see Figure 2 and Figure 3 , respectively give Figure 1 the preferred embodiment of the method applied to the update of the translation corpus and Figure 1 the preferred embodiment of the method applied to the update of the question and answer corpus.

[0074] Figure 2 A sustainable improved corpus problem classification method is given, which is based on a translation server cluster, the cluster includes N electronic devices {Dev1, Dev2, …, Dev N}, N>1; each of the electronic devices is configured with at least one translation engine;

[0075] The method comprises:

[0076] S210: i=1;

[0077] S220: Determine whether the electronic device Dev i generates translation corpus L i ; The translation corpus includes source corpus S i and target corpus O i ;

[0078] If yes, save the translation corpus L i to the pre-submission corpus and proceed to step S230;

[0079] If no, add the electronic device Dev i to the verification device set as a verification electronic device and proceed to step S230;

[0080] S230: Determine whether i < N holds. If yes, set i = i + 1 and return to step S220;

[0081] If no, proceed to step S240;

[0082] S240: Use the verification electronic devices in the verification device set to perform pre-submission operations on each translation corpus saved in the pre-submission corpus;

[0083] S250: Update the translation standard corpus according to the feedback results of the pre-submission operations. The update includes re-classifying and aligning or updating the alignment of the translation corpus in the translation standard corpus according to the source corpus and the target corpus.

[0084] Preferably, in the above Figure 2 embodiment, different verification electronic devices call different translation engines.

[0085] Next, a specific adaptation example is used to describe the implementation process of the above Figure 2 embodiment.

[0086] Assume that in the current translation scenario, there are 5 active user devices A1, A2, A3, A4, A5; among them, user device A2 is generating a translation corpus "magazine - magazine clip" to be submitted to the corpus to obtain a reward, and other users (A1, A3, A4, A5) are using the translation APP but not generating new corpus.

[0087] It can be understood that the regular translation corpus corresponding to the to-be-translated corpus "magazine" is usually "magazine" (already saved in the translation standard corpus), but in the mechanical field, especially in the field of special equipment, another translation of the to-be-translated corpus "magazine" also includes "magazine, magazine, magazine" and the like. Therefore, the translation corpus "magazine-magazine" generated by the user equipment A2 can be considered to be updated to the standard translation standard corpus, because the user did not first recommend the result when using the existing translation standard corpus.

[0088] At this time, the translation corpus "magazine-magazine" is saved to the pre-submission corpus first;

[0089] After 5 cycles of iterations, it is assumed that the translation corpus saved in the pre-submission corpus is only "magazine-magazine"; the verification equipment set is {A1, A3, A4, A5};

[0090] For the translation corpus, the verification electronic equipment of the verification equipment set performs a pre-submission operation on each translation corpus saved in the pre-submission corpus, and specifically includes:

[0091] The to-be-translated corpus corresponding to each translation corpus is used as the input of the translation engine of each verification electronic equipment, and the feedback result of the pre-submission operation is determined based on the comparison result of the output result of the translation engine and the post-translation corpus corresponding to each translation corpus.

[0092] Specifically, the translation engines of the translation APPs used by the other user equipment (A1, A3, A4, A5) are different, or correspond to different translation modes (translation fields), and in the presence of multiple other users, it is more likely that the translation result output by at least one user equipment is the same as or similar to the post-translation corpus corresponding to the translation corpus (magazine, magazine, magazine).

[0093] When the output result of the translation engine of at least one verification electronic equipment matches (i.e., the similarity is greater than a preset threshold) the comparison result of the post-translation corpus corresponding to the translation corpus, it is determined that the feedback result of the pre-submission operation is positive feedback.

[0094] At this time, according to the positive feedback result of the pre-submission operation, the translation standard corpus is updated, that is, "magazine" in the standard corpus is corresponding to "magazine, magazine, magazine", and "magazine" is updated to be classified as "mechanical-special equipment-gun".

[0095] As a preferred, for the translation corpus, the verification electronic equipment of the verification equipment set performs a pre-submission operation on each translation corpus saved in the pre-submission corpus, and specifically includes:

[0096] The post-translation corpus corresponding to each piece of translation corpus is taken as the input of the translation engine of each verification electronic device, and the feedback result of the pre-submission operation is determined based on the comparison result of the output result of the translation engine and the to-be-translated corpus corresponding to each piece of translation corpus.

[0097] Continuing the above example, the post-translation corpus (magazine, magazine, magazine) corresponding to the translation corpus "magazine- magazine" is taken as the input of the translation engine of the translation APP respectively used by other user devices (A1, A3, A4, A5), and the reverse translation result (magazine) is obtained;

[0098] At this time, only when the translation results output by the translation engines of the translation APPs respectively used by all other user devices (A1, A3, A4, A5) all contain the to-be-translated corpus corresponding to the translation corpus, the feedback result of the pre-submission operation is determined as positive feedback; as long as the translation result output by the translation engine of the translation APP used by one user device does not contain the to-be-translated corpus corresponding to the translation corpus at all, it is not necessary to submit, that is, the feedback result is negative feedback, and the operation of updating the standard corpus library is not performed.

[0099] At this time, the distributed atomic transaction operation characteristics of the distributed unit composed of all other user devices (A1, A3, A4, A5) are embodied, that is, the atomicity of "all execution or all non-execution" to ensure standard consistency.

[0100] Next, see Figure 3 , Figure 3 show Figure 1 The method is particularly applied to the preferred embodiment schematic diagram of the update of the question and answer corpus.

[0101] Similarly, Figure 3 The execution flow of the method is as follows:

[0102] A corpus problem classification method that can be continuously improved, the method is based on a question and answer server cluster, the cluster includes N electronic devices {Dev1, Dev2, …, Dev N N>1; each of the electronic devices is configured with at least one intelligent question and answer engine;

[0103] The method comprises:

[0104] S310: let i=1;

[0105] S320: judge whether the electronic device Dev i generates a question and answer corpus L i ; the question and answer corpus includes a question corpus S i and an answer corpus O i ;

[0106] If yes, the question-answer corpus L i saved to the pre-submission corpus, and step S330 is entered;

[0107] If no, the electronic device Dev i is added to the verification device set as a verification electronic device, and step S330 is entered;

[0108] S330: If yes, i is set to i+1, and step S320 is returned;

[0109] If no, step S340 is entered;

[0110] S340: The pre-submission operation is performed on each question-answer corpus saved in the pre-submission corpus by the verification electronic device in the verification device set;

[0111] S350: According to the feedback result of the pre-submission operation, the question-answer standard corpus is updated, and the update includes re-classification alignment or updating alignment of the question-answer corpus in the question-answer standard corpus according to the question corpus and the answer corpus.

[0112] As preferred, in the above Figure 2 embodiment, the intelligent question-answer engines called by different verification electronic devices are different.

[0113] In the intelligent customer service question-answer scenario Figure 3 , it is similar to the foregoing Figure 2 . The difference is that Figure 3 the question-answer corpus generated is usually in the form of “Q1-W1”, which is edited and maintained by user i, and needs to be determined for updating the standard question-answer corpus after verification, at which time, the other (N-1) users need to agree before it can be updated in the database. The agreement here means that if there is one disagreement, it cannot be stored in the database.

[0114] Specifically, the question corpus corresponding to each question-answer corpus is taken as the input of the intelligent question-answer engine of each verification electronic device, and the feedback result of the pre-submission operation is determined based on the comparison result of the output result of the intelligent question-answer engine and the answer corpus corresponding to each question-answer corpus.

[0115] Continuing the above example, it is assumed that in an APP scenario, there are 5 active user devices A1, A2, A3, A4, and A5; among them, the user device A2 edits and maintains a question-answer corpus in the form of “question 1-answer 1”, which is ready to be submitted to the corpus to obtain a reward, and the other users (A1, A3, A4, and A5) have not generated new corpora

[0116] At this time, the question-answer corpus “question 1-answer 1” is saved to the pre-submission corpus;

[0117] After 5 cycles of iterations, it is assumed that the pre-commit corpus saves only the question-answer corpus "question 1-answer 1"; the verification device set is {A1, A3, A4, A5}.

[0118] At this time, only when the four other devices {A1, A3, A4, A5} unanimously agree that the question-answer corpus "question 1-answer 1" can be saved, the feedback result of the pre-commit operation is a positive feedback result, and based on the positive feedback result, the operation of updating the standard corpus can be performed.

[0119] Therefore, the question corpus corresponding to each question-answer corpus is taken as the input of the intelligent question-answer engine of each verification electronic device, and when the output result of the intelligent question-answer engine of each verification electronic device is similar to the answer corpus corresponding to each question-answer corpus, the similarity is greater than a preset threshold, it is determined that the feedback result of the pre-commit operation is a positive feedback result, so as to embody atomicity and standard consistency.

[0120] Next, refer to Figure 4 , Figure 4 is Figure 1 a hardware / software architecture diagram of one specific implementation scenario of the method. The scenario includes a standard corpus, a cluster server and a plurality of electronic devices, each electronic device is configured with at least one AI engine, including an intelligent customer service engine, an intelligent question-answer engine, an intelligent translation engine, etc., and the corresponding standard corpus can be continuously improved and updated based on Figures 1-3 the sustainable improvement corpus question classification method.

[0121] Figure 5 is a functional module unit composition diagram of a sustainable improvement corpus question classification system according to an embodiment of the application.

[0122] The system includes a cluster server and a standard corpus, the cluster server is in wireless communication with N electronic devices, and the system further includes:

[0123] A corpus detection unit is configured to detect whether the electronic device generates parallel corpus;

[0124] A pre-commit unit is configured to save the parallel corpus generated by the electronic device to a pre-commit corpus;

[0125] A verification device specifying unit is configured to specify the electronic device as a verification electronic device when the electronic device does not generate parallel corpus;

[0126] A pre-operation execution unit is configured to perform a pre-commit operation on each parallel corpus saved in the pre-commit corpus by using the verification electronic device;

[0127] An updating unit is configured to update the standard corpus according to the feedback result of the pre-commit operation, and the updating includes reclassifying the parallel corpus of the standard corpus according to the source corpus or the target corpus.

[0128] The cluster server is a translation server and / or a question and answer server; the electronic device is configured with a translation engine or an intelligent question and answer engine; N is an integer greater than 2.

[0129] The verification device specifying unit forms a distributed transaction execution unit with all the verification electronic devices; the pre-commit operation of each parallel corpus saved in the pre-commit corpus by the distributed transaction execution unit forms a distributed atomic transaction operation.

[0130] In the distributed atomic transaction operation, a transaction is a reliable and independent work unit composed of a group of operations, and all the operations in the work unit are either all successful or all failed. A transaction has four basic characteristics: atomicity, consistency, isolation, and durability, also known as ACID characteristics of a transaction.

[0131] According to the foregoing method embodiments, the atomicity can be more clearly understood, that is:

[0132] Figure 2 In the preferred embodiment of the application, only when the translation results output by the translation engines of the translation APPs used by all the other user devices (A1, A3, A4, and A5) each respectively contain the to-be-translated corpus corresponding to the translation corpus, is the feedback result of the pre-commit operation determined to be positive feedback; as long as the translation result output by the translation engine of the translation APP used by one user device does not contain the to-be-translated corpus corresponding to the translation corpus at all, the pre-commit operation is not submitted, that is, the feedback result is negative feedback, and the operation of updating the standard corpus is not performed.

[0133] Figure 3 In the embodiment of the application, the question and answer corpus is usually in the form of "Q1-W1", and after being edited and maintained by user i, it needs to be determined whether to be recorded in the standard question and answer corpus for updating, at this time, the consensus of the other (N-1) users is required before being recorded in the corpus for updating. The consensus here means that as long as one user disagrees, the corpus cannot be recorded.

[0134] Therefore, the technical solution of the application can continuously improve and update the constructed corpus information, and the updating process can effectively avoid the corpus recording errors caused by the subjective bias accumulation of individual users or devices; in addition, the updating process is based on a distributed transaction processing mechanism, which ensures the standard consistency, and the advantages at least include:

[0135] (1) The user (device / terminal) who submits the corpus is separated from the user (device / terminal) who verifies the corpus: when a user (device / terminal) submits the corpus, the user (device / terminal) cannot serve as a verification device to achieve cross-validation.

[0136] (2) The corpus needs to be pre-submitted first to ensure standard consistency before formal submission.

[0137] (3) According to the feedback result of the pre-submission operation, the standard corpus library is updated, embodying the distributed atomic transaction operation.

[0138] Other technologies, principles, algorithms or models not detailed in the present application can be referred to the prior art.

[0139] In the foregoing embodiment section, the present application gives a plurality of embodiments, each of which can constitute an independent technical solution and can contribute to the prior art and solve the corresponding technical problems.

[0140] However, it should be pointed out that different embodiments can be combined with each other without violating the logic; at the same time, each embodiment can solve at least one technical problem, but it does not require each individual embodiment to solve multiple or all technical problems.

[0141] At the same time, in the specific embodiments of the present application, if user-related data is involved, when the embodiments of the present application are applied to specific products or technologies, the user's permission or consent needs to be obtained, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0142] The method embodiments and systems of the present application have been shown and described, but for those of ordinary skill in the art, it can be understood that various changes, modifications, replacements and variations can be made to these embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method of sustainable improved corpus problem classification, said method based on a cluster implementation, said cluster comprising N electronic devices {Dev1, Dev2,..., DevN}, N > 1; each of said electronic devices configured with at least one translation engine; N} characterized in that The method comprises the following steps: S210: let i = 1; S220: judging whether the electronic device Dev i generates a translation corpus L i ; the translation corpus includes a to-be-translated corpus S i and a post-translation corpus O i ; If yes, the translation corpus L i is saved to the pre-submission corpus and step S230 is entered. If no, the electronic device Dev i As the electronic device joins the set of verification devices, step S230 is entered. S230: determine whether i < N is true, if yes, let i = i + 1, return to step S220; if no, enter step S240; S240: using the verification electronic device of the verification device set, pre-submits each translation corpus saved by the pre-submission corpus; S250: when the feedback result of the pre-submission operation is positive feedback, updating the translation standard corpus, the updating comprises reclassifying and aligning or updating the alignment of the translation corpus of the translation standard corpus according to the to-be-translated corpus and the translated corpus; Wherein, step S240 performs pre-submission operation on each translation corpus saved by the pre-submission corpus, specifically comprising: Taking the translated corpus corresponding to each translation corpus as the input of the translation engine of each verification electronic device; When the translation results output by the translation engines of the translation APPs used by each verification electronic device respectively all contain the to-be-translated corpus corresponding to the translation corpus, it is determined that the feedback result of the pre-submission operation is positive feedback.

2. The method of claim 1, wherein, The to-be-translated corpus and the translated corpus are different languages.

3. The method of claim 1, wherein, The translation engines of the translation APPs used by each verification electronic device respectively are different or correspond to different translation modes.

4. The method of claim 1, wherein the corpus is continually improved. The cluster is a translation server cluster; the cluster translation server is connected to the standard translation corpus; The user inputs the to-be-translated word W1 or to-be-translated sentence S1 of the first language as the to-be-translated corpus, and calls the standard translation corpus through the cluster translator server to execute the translation engine and output the corresponding translation result as the translated word W2 or translated sentence S2 of the second language as the translated corpus.

5. A method of sustainable improvement of corpus problem classification, the method is based on cluster implementation, the cluster includes N electronic devices {Dev1, Dev2, …, DevN}, N > 1; each of the electronic devices is configured with at least one intelligent question and answer engine. N} The method comprises: S310: let i = 1; S320: determining whether the electronic device Dev i generates the question-answer corpus L i ; the question-answer corpus includes question corpus S i and answer corpus O i ; If yes, the question-answer corpus L is asked i saved to the pre-submission corpus, and step S330 is entered. If no, the electronic device Dev i As the electronic device joins the set of verification devices, step S330 is entered. S330: determine whether i < N is true, if yes, let i = i + 1, return to step S320; if no, enter step S340; S340: using the verification electronic device of the verification device set, pre-submits each question and answer corpus saved by the pre-submission corpus; S350: when the feedback result of the pre-submission operation is positive feedback, updating the question and answer standard corpus, the updating comprises reclassifying and aligning or updating the alignment of the question and answer corpus of the question and answer standard corpus according to the question corpus and the answer corpus; Wherein, the step S340 specifically comprises: Taking the question corpus corresponding to each question and answer corpus as the input of the intelligent question and answer engine of each verification electronic device, when the output results of the intelligent question and answer engine of each verification electronic device are all greater than a preset threshold value from the similarity of the answer corpus corresponding to each question and answer corpus, it is determined that the feedback result of the pre-submission operation is positive feedback.

6. A method of sustainable improvement of corpus problem classification as claimed in claim 5, wherein, The question corpus and the answer corpus are the same language.

7. A method of sustainable improvement of corpus problem classification as claimed in claim 5 wherein, The cluster is a question and answer server cluster; each electronic device is configured with an intelligent question and answer engine; the intelligent question and answer engines called by different verification electronic devices are different each time.

8. A sustainable improvement corpus problem classification system, characterized by, The system comprises a cluster server and a standard corpus, the cluster server is in wireless communication with N electronic devices, and is used for realizing the sustainable improvement of the corpus problem classification method according to any one of claims 1-7.

9. A continually improvable corpus question classification system as claimed in claim 8, wherein, The system comprises a verification device designation unit, which designates an electronic device as a verification electronic device when the electronic device does not generate parallel corpus; All the verification electronic devices constitute a distributed transaction execution unit; the pre-commit operation of each parallel corpus by the distributed transaction execution unit constitutes a distributed atomic transaction operation; the parallel corpus is a translation corpus or a question and answer corpus.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer executable instructions, and the computer executable instructions are executed by the processor to implement the steps of the sustainable improved corpus problem classification method in any one of claims 1 to 7.

Images