CN116049362A

CN116049362A - Big data optimization method and server applied to digital cloud service

Info

Publication number: CN116049362A
Application number: CN202211670599.0A
Authority: CN
Inventors: 马越扬; 薛念慈
Original assignee: Hefei Zhaofal Internet Technology Co ltd
Current assignee: Hefei Zhaofal Internet Technology Co ltd
Priority date: 2022-12-26
Filing date: 2022-12-26
Publication date: 2023-05-02

Abstract

The invention provides a big data optimization method and a server applied to digital cloud service, which do not need repeated extraction and mining of consultation response characteristics when determining the big data of the final consultation response text based on a consultation response characteristic analysis strategy, so that the efficiency of the big data of the final consultation response text can be improved, and unnecessary calculation load waste is reduced. Further, by carrying out demand analysis on the final consultation response text big data, the software service optimization demand can be accurately obtained, and the optimization updating of the software service can be realized by combining the preset resource use constraint condition. In view of the fact that the resource use constraint condition is used for achieving overload protection of the target service platform system, the problem of system overload can be considered when the target service platform system is instructed to conduct software service optimization updating, excessive use of system resources caused by starting a large amount of optimization updating tasks in a short time is avoided, and abnormal operation of the target service platform system is further avoided.

Description

Big data optimization method and server applied to digital cloud service

Technical Field

The invention relates to the technical field of big data, in particular to a big data optimization method and a server applied to digital cloud service.

Background

Today, big data and artificial intelligence fire explosion accelerate the pulsation of the cloud computing market, and an intelligent technology represented by AI big data is incorporated into the layout of cloud services. The digitalized and intelligent cloud service brings a brand new development mode for various industries. For digital/intelligent cloud services, focusing on listening to user needs is a viable way. However, it is not preferable to ignore overload of the service system too pursuing to meet the user's needs, but it is still difficult for the conventional technology to cross this barrier.

Disclosure of Invention

The invention provides a big data optimization method and a server applied to digital cloud service, and the technical scheme is as follows.

The first aspect is a big data optimization method applied to a digital cloud service, applied to a big data optimization server, the method comprising:

when receiving a consultation response text query application, determining final consultation response text big data by utilizing a consultation response characteristic analysis strategy; the final consultation response text big data is consultation response text big data of the target digital cloud service;

obtaining software service optimization requirements aiming at the target digital cloud service according to the final consultation response text big data;

Based on the software service update requirement and a preset resource use constraint condition, optimizing and updating the software service for the target digital cloud service; the resource use constraint condition is used for realizing overload protection of a target service platform system, and the target service platform system is a platform system corresponding to the target digital cloud service.

In some alternative embodiments, the determining the final counseling response text big data using the counseling response feature analysis strategy includes:

acquiring initial consultation response text big data, initial consultation response frequent item knowledge corresponding to the initial consultation response text big data and at least one group of supplementary target consultation response text big data, wherein the target consultation response text big data is supplementary consultation response text big data for the initial consultation response text big data with a query error;

carrying out consultation response vector mining on the target consultation response text big data to obtain target consultation response frequent knowledge corresponding to the target consultation response text big data;

the initial consultation response frequent item knowledge is adjusted by utilizing the target consultation response frequent item knowledge to obtain adjusted consultation response frequent item knowledge, wherein the adjusted consultation response frequent item knowledge and the target consultation response frequent item knowledge are consultation response frequent item knowledge under the same knowledge network;

Carrying out knowledge collection on the target consultation response frequent knowledge and the adjusted consultation response frequent knowledge to obtain linkage consultation response frequent knowledge for inquiring consultation response text;

responding to a consultation response text query application, carrying out consultation response vector mining on the large data of the consultation response text to be queried contained in the consultation response text query application, and determining final large data of the consultation response text in the large data of the initial consultation response text and the large data of the target consultation response text according to the determined frequent knowledge of the consultation response to be queried and the linked frequent knowledge of the consultation response.

In some optional embodiments, the adjusting the initial frequent counseling knowledge by using the target frequent counseling knowledge to obtain adjusted frequent counseling knowledge includes:

acquiring initial knowledge network data of the initial consultation response frequent knowledge and target knowledge network data of the target consultation response frequent knowledge, and comparing the initial knowledge network data with the target knowledge network data;

and through comparing the results, transforming the initial frequent consultation response knowledge to the knowledge network of the target frequent consultation response knowledge by adopting a knowledge transformation layer of the debugged frequent consultation response text query algorithm to obtain the regulated frequent consultation response knowledge.

In some optional embodiments, the step of transforming the initial frequent query response knowledge to the knowledge network of the target frequent query response knowledge by using the knowledge transformation layer of the debugged frequent query response text query algorithm through the comparison result, before obtaining the adjusted frequent query response knowledge, further includes:

acquiring a consultation response reference text record, wherein the consultation response reference text record comprises an initial consultation response reference text set and a supplementary target consultation response reference text set;

determining consultation response reference joint text for debugging in the initial consultation response reference text set and the target consultation response reference text set;

debugging a first consultation response text query layer of a general consultation response text query algorithm by using the initial consultation response reference text set and the consultation response reference joint text to obtain a debugged first consultation response text query layer;

and based on the consultation response reference text record and the consultation response reference joint text, performing collaborative debugging on the first consultation response text query layer after debugging and a knowledge transformation layer in the general consultation response text query algorithm to obtain a debugged consultation response text query algorithm.

In some alternative embodiments, the obtaining the consulting response reference text record includes:

acquiring an initial consultation response reference text set, acquiring big consultation response text data with a query error of a second consultation response text query layer in the general consultation response text query algorithm, and obtaining a consultation response noise text set, wherein the second consultation response text query layer is obtained by debugging the initial consultation response reference text set;

obtaining derived text of the initial consultation response reference text set, obtaining a consultation response derived text set, and taking the consultation response noise text set and the consultation response derived text set as complementary target consultation response reference text sets;

and carrying out knowledge collection on the target consultation response reference text set and the initial consultation response reference text set to obtain the consultation response reference text record.

In some optional embodiments, the determining the consultation response reference joint text for debugging in the initial consultation response reference text set and the target consultation response reference text set includes:

extracting a target consultation response reference text from the initial consultation response reference text set and the target consultation response reference text set;

Respectively determining the text difference degree between the target consultation response reference text and the rest consultation response reference text in the consultation response reference text record;

and based on the text difference degree, extracting a set number of negative examples corresponding to the target consultation response reference text from the consultation response reference text record, and adding the negative examples to the consultation response reference text record corresponding to the target consultation response reference text to obtain the consultation response reference joint text of the target consultation response reference text.

In some optional embodiments, the debugging the first consultation response text query layer of the general consultation response text query algorithm by using the initial consultation response reference text set and the consultation response reference joint text to obtain a debugged first consultation response text query layer includes:

adopting a first consultation response text query layer of a general consultation response text query algorithm to carry out consultation response vector mining on initial consultation response reference texts in the initial consultation response reference text set, and obtaining the initial consultation response frequent item reference knowledge;

extracting target initial consultation response frequent knowledge corresponding to the consultation response reference joint text from the initial consultation response frequent knowledge;

Determining a knowledge vector difference value between the target initial consultation response frequent knowledge items to obtain initial joint debugging cost corresponding to the consultation response reference joint text;

and optimizing the first consultation response text query layer based on the initial joint debugging cost to obtain a first consultation response text query layer after debugging.

In some optional embodiments, the collaborative debugging is performed on the first after-debugging consultation response text query layer and the knowledge transformation layer in the general consultation response text query algorithm based on the consultation response reference text record and the consultation response reference joint text, to obtain a debugged consultation response text query algorithm, including:

adopting the debugged first consultation response text query layer to carry out consultation response vector mining on the consultation response reference text in the consultation response reference text record, so as to obtain first consultation response frequent knowledge;

adopting the second consultation response text query layer to carry out consultation response vector mining on the consultation response reference text in the consultation response reference text record, and adopting a knowledge transformation layer in the general consultation response text query algorithm to transform the mined frequent knowledge to obtain second consultation response frequent knowledge;

And optimizing the debugged first consultation response text query layer and the knowledge transformation layer by utilizing the first consultation response frequent knowledge, the second consultation response frequent knowledge and the consultation response reference joint text to obtain a debugged consultation response text query algorithm.

In some optional embodiments, the optimizing the first post-debugging consultation response text query layer and the knowledge transformation layer to obtain a debugged consultation response text query algorithm by using the first consultation response frequent knowledge, the second consultation response frequent knowledge and the consultation response reference joint text includes:

extracting a combined sample set formed by the consultation response reference texts in the target consultation response reference text set from the consultation response reference text set to obtain a target consultation response combined text;

extracting a combined sample set consisting of the consultation response reference text in the target consultation response reference text set and the consultation response reference text in the initial consultation response reference text set from the consultation response reference combined text to obtain a global consultation response reference text;

determining target debugging cost data corresponding to the consultation response reference text record by utilizing the target consultation response combined text, the global consultation response reference text, the first consultation response frequent item knowledge and the second consultation response frequent item knowledge;

And optimizing the first consultation response text query layer and the knowledge transformation layer after debugging by using the target debugging cost data to obtain a debugged consultation response text query algorithm.

In some optional embodiments, the determining, by using the target consultation response joint text, the global consultation response reference text, the first frequent knowledge of consultation responses, and the second frequent knowledge of consultation responses, the target debug cost data corresponding to the consultation response reference text record includes:

determining the joint debugging cost of the target consultation response joint text according to the first consultation response frequent knowledge to obtain the target joint debugging cost;

based on the second consultation response frequent knowledge, determining the joint debugging cost of the global consultation response reference text to obtain global joint debugging cost;

and carrying out knowledge collection on the target joint debugging cost and the global joint debugging cost to obtain target debugging cost data corresponding to the consultation response reference text record.

In some optional embodiments, the determining, based on the second frequent knowledge of the consultation response, a joint debugging cost of the global consultation response reference text, to obtain a global joint debugging cost includes:

Extracting any one consultation response reference text from the global consultation response reference text as a current consultation response reference text for determining joint debugging cost;

determining the joint debugging cost of the global consultation response reference text by utilizing the current consultation response reference text and the second consultation response frequent knowledge to obtain basic global joint debugging cost;

jumping to the step of extracting any consultation response reference text from the global consultation response reference text as the current consultation response reference text for determining the joint debugging cost until all the consultation response reference texts in the global consultation response reference text are used as the current consultation response reference text for determining the joint debugging cost, and obtaining the basic global joint debugging cost corresponding to each consultation response reference text in the global consultation response reference text;

and carrying out knowledge collection on the basic global joint debugging cost to obtain the global joint debugging cost.

In some optional embodiments, the determining final big data of the consultation response text in the big data of the initial consultation response text and the big data of the target consultation response text according to the determined frequent knowledge of the consultation response to be queried and the linked frequent knowledge of the consultation response, includes:

Determining a knowledge vector difference value between the to-be-queried consultation response frequent knowledge and each consultation response frequent knowledge in the linkage consultation response frequent knowledge to obtain a first knowledge vector difference value;

and finishing the first knowledge vector difference value, and determining final consultation response text big data from the initial consultation response text big data and the target consultation response text big data according to the finishing result.

The second aspect is a big data optimization server comprising a memory and a processor; the memory is coupled to the processor; the memory is used for storing computer program codes, and the computer program codes comprise computer instructions; wherein the computer instructions, when executed by the processor, cause the big data optimization server to perform the method of the first aspect.

A third aspect is a computer readable storage medium having stored thereon a computer program which, when run, performs the method of the first aspect.

According to one embodiment of the invention, when a consultation response text query application is received, final consultation response text big data is determined by utilizing a consultation response feature analysis strategy; the final consultation response text big data is consultation response text big data of the target digital cloud service; obtaining software service optimization requirements aiming at the target digital cloud service according to the final consultation response text big data; based on the software service update requirement and a preset resource use constraint condition, optimizing and updating the software service for the target digital cloud service; the resource use constraint condition is used for realizing overload protection of a target service platform system, and the target service platform system is a platform system corresponding to the target digital cloud service.

It can be understood that when the final consultation response text big data is determined based on the consultation response feature analysis strategy, repeated consultation response feature extraction and mining are not required, so that the efficiency of the final consultation response text big data can be improved, and unnecessary calculation load waste is reduced. Further, by carrying out demand analysis on the final consultation response text big data, the software service optimization demand can be accurately obtained, and the optimization updating of the software service can be realized by combining the preset resource use constraint condition. In view of the fact that the resource use constraint condition is used for achieving overload protection of the target service platform system, the problem of system overload can be considered when the target service platform system is instructed to conduct software service optimization updating, excessive use of system resources caused by starting a large amount of optimization updating tasks in a short time is avoided, and abnormal operation of the target service platform system is further avoided.

Further, after initial consultation response text big data, initial consultation response frequent item knowledge corresponding to the initial consultation response text big data and at least one group of supplementary target consultation response text big data are obtained, consultation response vector mining is carried out on the target consultation response text big data to obtain target consultation response frequent item knowledge corresponding to the target consultation response text big data, then the initial consultation response frequent item knowledge is adjusted according to the target consultation response frequent item knowledge to obtain adjusted consultation response frequent item knowledge, then the target consultation response frequent item knowledge and the adjusted consultation response frequent item knowledge are subjected to knowledge collection to obtain linkage consultation response frequent item knowledge for consultation response text query, consultation response vector mining is carried out on the to-be-queried consultation response text big data contained in the consultation response text query application, and final consultation response text big data is determined in the initial consultation response text big data and the target consultation response text big data according to the determined to-be-queried consultation response frequent item knowledge; in view of the fact that the embodiment of the invention can transform the initial consultation response frequent knowledge into the knowledge network of the target consultation response frequent knowledge, the consultation response frequent knowledge for inquiring the consultation response text can be timely adjusted, the initial consultation response text big data does not need to be subjected to consultation response vector mining again, the operation load required by the consultation response vector mining is effectively saved, and the mining timeliness of the consultation response frequent knowledge can be improved, so that the timeliness of inquiring the consultation response text is ensured.

Drawings

Fig. 1 is a flow chart of a big data optimization method applied to a digital cloud service according to an embodiment of the present invention.

Detailed Description

Hereinafter, the terms "first," "second," and "third," etc. are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first", "a second", or "a third", etc., may explicitly or implicitly include one or more such feature.

Fig. 1 shows a flow chart of a big data optimization method applied to a digital cloud service, which is provided by an embodiment of the present invention, and the big data optimization method applied to the digital cloud service may be implemented by a big data optimization server, where the big data optimization server may include a memory and a processor; the memory is coupled to the processor; the memory is used for storing computer program codes, and the computer program codes comprise computer instructions; when the processor executes the computer instructions, the big data optimization server is caused to execute the technical schemes described in the first to third steps.

And step one, when a consultation response text query application is received, determining final consultation response text big data by utilizing a consultation response characteristic analysis strategy.

And the final consultation response text big data is the consultation response text big data of the target digital cloud service.

In the embodiment of the invention, the big data optimization server can receive the consultation response text query application aiming at the target digital cloud service, and then query based on the consultation response characteristic analysis strategy capable of improving query timeliness to obtain final consultation response text big data. Further, the target digital cloud service may be an e-commerce service, a supply chain financial service, an enterprise business service, or the like.

And step two, obtaining the software service optimization requirement aiming at the target digital cloud service according to the final consultation response text big data.

It can be understood that the large consultation response text data of the target digital cloud service includes consultation response information between users or between users and service providers, and the consultation response information can reflect the requirements of the users for the target digital cloud service. The software service optimization requirements comprise an interface operation guided optimization requirement, an upgrade requirement of personal information protection, an interception mechanism adjustment requirement of data pushing and the like.

And thirdly, optimizing and updating the software service for the target digital cloud service based on the software service updating requirement and a preset resource use constraint condition.

The resource use constraint condition is used for realizing overload protection of a target service platform system, and the target service platform system is a platform system corresponding to the target digital cloud service. It can be understood that when the optimization updating of the software service is performed on the target digital cloud service, the software service updating requirement and the resource usage constraint condition need to be considered simultaneously, for example, the system resource proportion occupied by the optimization updating task M1 corresponding to the optimization requirement guided by the interface operation is 10%, the idle resource proportion of the target service platform system is 15%, and the resource usage constraint condition is that the minimum margin of idle resources is 10%. In this case, the big data optimization server may analyze that if the optimization update task M1 corresponding to the optimization requirement guided by the interface operation is executed, the idle resource ratio of the target service platform system may be reduced to 5%, and the resource usage constraint condition is not satisfied, so the big data optimization server may not issue the optimization update task M1 corresponding to the optimization requirement guided by the interface operation to the target service platform system, and may wait for the decision analysis of whether to execute the optimization update after the idle resource ratio of the target service platform system increases.

For another example, the interception mechanism of data push adjusts the system resource proportion occupied by the required optimization update task M2 to be 5%, the idle resource proportion of the target service platform system to be 15%, and the resource use constraint condition to be the minimum margin of the idle resource to be 10%. In this case, the big data optimization server may analyze that if the interception mechanism of data push adjusts the required optimization update task M2, the idle resource ratio of the target service platform system may be reduced to 10%, so as to satisfy the constraint condition of resource usage, so that the big data optimization server may issue the interception mechanism of data push to the target service platform system to adjust the required optimization update task M2, and thus, the target service platform system may perform targeted optimization update of the software service on the premise of ensuring that the target service platform system is not overloaded.

When the method is applied to the first to third steps, the repeated extraction and mining of the consultation response characteristics are not needed when the final consultation response text big data is determined based on the consultation response characteristic analysis strategy, so that the efficiency of the final consultation response text big data can be improved, and unnecessary calculation load waste is reduced. Further, by carrying out demand analysis on the final consultation response text big data, the software service optimization demand can be accurately obtained, and the optimization updating of the software service can be realized by combining the preset resource use constraint condition. In view of the fact that the resource use constraint condition is used for achieving overload protection of the target service platform system, the problem of system overload can be considered when the target service platform system is instructed to conduct software service optimization updating, excessive use of system resources caused by starting a large amount of optimization updating tasks in a short time is avoided, and abnormal operation of the target service platform system is further avoided.

Under some possible design ideas, the determining of the final consultation response text big data by utilizing the consultation response feature analysis strategy described in the first step may include the technical solutions described in S101-S105.

S101, acquiring initial consultation response text big data, initial consultation response frequent item knowledge corresponding to the initial consultation response text big data and at least one group of supplementary target consultation response text big data.

The target large data of the consultation response text is large data of the consultation response text which is supplemented for the occurrence of the query error of the large data of the initial consultation response text (the large data of the original consultation response text), the occurrence of the query error of the large data of the initial consultation response text can be understood as that the large data of the consultation response text which needs to be queried is not queried in the large data of the initial consultation response text or the large data of the incorrect consultation response text is determined, based on the large data, one or more groups of the large data of the target consultation response text needs to be newly added in a data set formed by the large data of the initial consultation response text, and the large data of the target consultation response text is the large data of the consultation response text which needs to be queried is in order to enable the data set to exist, so that the query precision of the consultation response text is improved.

The concept of acquiring the initial consultation response text big data, the initial consultation response frequent knowledge (the original response interaction text feature) and the target consultation response text big data can be various, and examples can include the following related contents.

For example, for the initial consultation response text big data and the initial consultation response frequent knowledge, the initial consultation response text big data and the initial consultation response frequent knowledge can be directly obtained, or the initial consultation response text big data can also be obtained, and the initial consultation response text big data is subjected to consultation response vector mining (feature mining or feature extraction) by adopting an initial consultation response text query algorithm, so that the initial consultation response frequent knowledge corresponding to the initial consultation response text big data is obtained.

For the target large data of the consultation response text, the initial consultation response text query algorithm can be adopted to carry out consultation response vector mining on the initial large data of the consultation response text, so as to obtain initial consultation response frequent item knowledge corresponding to the initial large data of the consultation response text, the target large data of the consultation response text to be queried is input into the initial consultation response text query algorithm, the alternative large data of the consultation response text is determined in the initial large data of the consultation response text according to the initial consultation response frequent item knowledge by the initial consultation response text query algorithm, when the determined alternative large data of the consultation response text is abnormal or has errors, the target large data of the consultation response text to be queried is used as the target large data of the consultation response text to be supplemented, or the consultation response type of the initial large data of the consultation response text can be obtained, the large data of the consultation text different from the consultation response type is extracted in the set data set according to the consultation response type as the derivative text (predictive text) of the initial large data of the consultation response text, or the derivative text is also received and the large data of the consultation text to be added to be required to the initial large data of the consultation text to be updated in the initial large data of the consultation text response text to be required to be supplemented, so as to obtain the target large data of the consultation text.

S102, carrying out consultation response vector mining on the target consultation response text big data to obtain target consultation response frequent knowledge corresponding to the target consultation response text big data.

For example, the first consultation response text query layer of the debugged consultation response text query algorithm can be adopted to perform consultation response vector mining on the target consultation response text big data, so that the target consultation response frequent knowledge corresponding to the target consultation response text big data is obtained.

The first consultation response text query layer for performing consultation response vector mining on the target consultation response text big data is different from the consultation response text query layer for extracting initial consultation response frequent knowledge of the initial consultation response text big data, and the initial consultation response frequent knowledge can be obtained by performing consultation response vector mining on the initial consultation response text big data by the initial consultation response text query algorithm in the big data optimization server or the second consultation response text query layer in the general consultation response text query algorithm. The model architecture of the first consultation response text query layer can be various, for example, a deep learning model.

And S103, adjusting the initial consultation response frequent knowledge according to the target consultation response frequent knowledge to obtain the adjusted consultation response frequent knowledge.

The adjusted frequent consultation response knowledge and the target frequent consultation response knowledge are the frequent consultation response knowledge under the same knowledge network, so that the adjustment can be understood as transforming (mapping) the initial frequent consultation response knowledge to the knowledge network (feature vector space) of the target frequent consultation response knowledge, so that the target frequent consultation response knowledge and the adjusted frequent consultation response knowledge can mutually perform frequent consultation, and the frequent consultation response knowledge in the frequent consultation response knowledge base of the big data optimization server can be formed.

The concept of adjusting the initial consultation response frequent knowledge can be various, and examples can include the following related content.

For example, the initial knowledge network data of the initial consultation response frequent knowledge and the target knowledge network data of the target consultation response frequent knowledge can be obtained, the initial knowledge network data and the target knowledge network data are compared, and the initial consultation response frequent knowledge is transformed into the knowledge network of the target consultation response frequent knowledge by adopting a knowledge transformation layer (feature mapping model) of the debugged consultation response text query algorithm through the comparison result, so that the regulated consultation response frequent knowledge is obtained.

The concept of transforming the initial frequent consultation response knowledge to the knowledge network of the target frequent consultation response knowledge by using the knowledge transformation layer through the comparison result may be various, for example, an indication variable (mapping parameter) for transforming the initial frequent consultation response knowledge is determined through the comparison result, based on the indication variable, the initial frequent consultation response knowledge is transformed to the knowledge network of the target frequent consultation response knowledge by using the knowledge transformation layer, so as to obtain the adjusted frequent consultation response knowledge, or an indication variable corresponding to the comparison result is extracted from the indication variable set of the knowledge transformation layer through the comparison result, so as to obtain a target indication variable, and based on the target indication variable, the initial frequent consultation response knowledge is transformed to the knowledge network of the target frequent consultation response knowledge by using the knowledge transformation layer, so as to obtain the adjusted frequent consultation response knowledge.

The method for optimizing big data applied to the digital cloud service before the step of adopting the knowledge transformation layer of the debugged consultation response text query algorithm to transform the initial consultation response frequent knowledge to the knowledge network of the target consultation response frequent knowledge to obtain the adjusted consultation response frequent knowledge can further comprise the following steps: obtaining a consultation response reference text record (consultation response text training data), wherein the consultation response reference text record comprises an initial consultation response reference text set (original consultation response text training pair) and a supplementary target consultation response reference text set (target consultation response text training pair), determining consultation response reference joint text (which can be understood as consultation response text triples) for debugging in the initial consultation response reference text set and the target consultation response reference text set, debugging a first consultation response text query layer of the general consultation response text query algorithm according to the initial consultation response reference text set and the consultation response reference joint text, obtaining a first consultation response text query layer after debugging, and performing collaborative debugging (which can be understood as joint training) on knowledge transformation layers in the first consultation response text query layer after debugging and the general consultation response text query algorithm based on the consultation response reference text record and the consultation response reference joint text, so as to obtain a debugged consultation response text query algorithm.

The concept of acquiring the reference text record of the counseling response may be various, for example, the initial reference text of the counseling response may be acquired, and the big data of the counseling response text with the query error in the second counseling response text query layer in the general counseling response text query algorithm may be acquired, so as to obtain the noise text set of the counseling response, where the second counseling response text query layer is debugged from the initial reference text set of the counseling response. And obtaining derived texts of the initial consultation response reference text set, obtaining a consultation response derived text set, and taking the consultation response noise text set and the consultation response derived text set as complementary target consultation response reference text sets.

After the initial consultation response reference text is acquired, the consultation response reference joint text for debugging can be determined in the initial consultation response reference text set and the target consultation response reference text set, and the consultation response reference joint text can be a consultation response reference text set formed by a positive example and a negative example of the consultation response reference text corresponding to the consultation response reference text. The concept of extracting the combined consultation response reference text can be various, for example, a target consultation response reference text can be extracted from an initial consultation response reference text set and a target consultation response reference text, the text difference degree between the target consultation response reference text and the rest consultation response reference texts in the consultation response reference text records is respectively determined, a set number of negative examples corresponding to the target consultation response reference text are extracted from the consultation response reference text records based on the text difference degree, and the consultation response reference text records corresponding to the target consultation response reference text in the negative examples are added to obtain the combined consultation response reference text of the target consultation response reference text.

In the embodiment of the invention, the constitution ideas of the consultation response reference text in the consultation response reference joint text can comprise three types: firstly, an initial consultation response reference text set+a certain group of consultation response reference texts in other initial consultation response reference text sets, secondly, a target consultation response reference text set+a certain group of consultation response reference texts in other target consultation response reference text sets, and thirdly, the initial consultation response reference text set+a certain group of consultation response reference texts in the target consultation response reference text sets/the target consultation response reference text sets+a certain group of consultation response reference texts in the initial consultation response reference text sets.

After determining the consultation response reference joint text used for debugging, the first consultation response text query layer of the general consultation response text query algorithm (untrained) can be debugged based on the consultation response reference joint text to obtain a debugged first consultation response text query layer, and various debugging ideas can be provided, for example, the first consultation response text query layer of the general consultation response text query algorithm is adopted to carry out consultation response vector mining on the initial consultation response reference text in the initial consultation response reference text set to obtain initial consultation response frequent item reference knowledge, target initial consultation response frequent item knowledge corresponding to the consultation response reference text is extracted from the initial consultation response frequent item reference knowledge, a knowledge vector difference value between the target initial consultation response frequent item knowledge is calculated to obtain initial joint debugging cost corresponding to the consultation response reference joint text, and the first consultation response text query layer is optimized based on the initial joint debugging cost to obtain the debugged first consultation response text query layer.

When the first consultation response text query layer is debugged, a complete initial consultation response reference text set is input, so that the type of the consultation response reference combined text corresponding to the target initial consultation response frequent knowledge to be extracted can be a combined sample set consisting of the initial consultation response reference text set and a certain consultation response reference text in other initial consultation response reference text sets, and can also be called as an initial consultation response combined text. Therefore, the concept of extracting the target initial consultation response frequent item knowledge corresponding to the initial consultation response joint text from the initial consultation response frequent item knowledge may be various, for example, the target initial sample corresponding to the consultation response frequent item knowledge in the initial consultation response joint text may be extracted from the initial consultation response frequent item reference knowledge to obtain the first initial consultation response frequent item knowledge, the consultation response frequent item knowledge corresponding to the positive example of the target initial consultation response reference text is extracted from the initial consultation response frequent item reference knowledge to obtain the second initial consultation response frequent item knowledge, the consultation response frequent item knowledge corresponding to the negative example of the target initial consultation response reference text is extracted from the initial consultation response frequent item knowledge to obtain the third initial consultation response frequent item knowledge, and the first initial consultation response frequent item knowledge, the second initial consultation response frequent item knowledge and the third initial consultation response frequent item knowledge are used as the target initial consultation response frequent item knowledge.

After the knowledge of the target initial consultation response frequent item is extracted, a knowledge vector difference value between the knowledge of the target initial consultation response frequent item can be determined, initial joint debugging cost corresponding to the consultation response reference joint text is obtained, multiple ideas for calculating the initial joint debugging cost can be provided, for example, the knowledge vector difference value of the first initial consultation response frequent item knowledge and the knowledge of the second initial consultation response frequent item knowledge can be determined, a first knowledge vector difference value is obtained, the knowledge vector difference value of the first initial consultation response frequent item knowledge and the knowledge of the third initial consultation response frequent item knowledge is calculated, a second knowledge vector difference value is obtained, a comparison value of the first knowledge vector difference value and the second knowledge vector difference value is calculated, and the comparison value is fused with a preset difference value, so that the joint comparison value is obtained. And comparing the joint comparison value with a preset difference threshold, when the joint comparison value exceeds the preset difference threshold, taking the joint comparison value as an initial joint debugging cost, and when the joint comparison value does not exceed the preset difference threshold, taking the preset difference threshold as the initial joint debugging cost (a triplet loss function).

It can be understood that after the first consultation response text query layer is debugged, the first consultation response text query layer after the debugging and the knowledge transformation layer in the general consultation response text query algorithm can be cooperatively debugged based on the consultation response reference text record and the consultation response reference joint text, and various ideas of the collaborative debugging can be adopted, for example, the first consultation response text query layer after the debugging can be adopted to carry out consultation response vector mining on the consultation response reference text in the consultation response reference text record to obtain first consultation response frequent item knowledge, the second consultation response text query layer is adopted to carry out consultation response vector mining on the consultation response reference text in the consultation response reference text record to obtain second consultation response frequent item knowledge, and the first consultation response text query layer and the knowledge transformation layer after the debugging are optimized according to the first consultation response frequent item knowledge, the second consultation response frequent item knowledge and the consultation response joint text, so as to obtain the debugged consultation response text query algorithm.

The concept of optimizing the debugged first consultation response text query layer and knowledge transformation layer according to the first consultation response frequent knowledge, the second consultation response frequent knowledge and the consultation response reference joint text can be various, for example, a joint sample set formed by the consultation response reference texts in the target consultation response reference text set can be extracted from the consultation response reference joint text to obtain a target consultation response joint text, a joint sample set formed by the consultation response reference texts in the target consultation response reference text set and the consultation response reference texts in the initial consultation response reference text set is extracted from the consultation response reference joint text to obtain a global consultation response reference text, and the debugged consultation response text query algorithm is obtained by optimizing the debugged first consultation response text query layer and the knowledge transformation layer according to the target debugging cost data according to the target consultation response reference text record.

The target consultation response combined text may be composed of three consultation response reference texts in the target consultation response reference text set, and the global consultation response reference text may be composed of a consultation response reference text pair in the target consultation response reference text set and the initial consultation response reference text set, for example, may include two consultation response reference texts in the target consultation response reference text set and one consultation response reference text in the initial consultation response reference text set, or may further include two consultation response reference texts in the initial consultation response reference text set and one consultation response reference text in the target consultation response reference text set. The concept of determining the target debugging cost data corresponding to the consultation response reference text record can be multiple according to the target consultation response joint text, the global consultation response reference text, the first consultation response frequent item knowledge and the second consultation response frequent item knowledge, for example, the joint debugging cost of the target consultation response joint text is determined according to the first consultation response frequent item knowledge to obtain the target joint debugging cost, the joint debugging cost of the global consultation response reference text is determined based on the second consultation response frequent item knowledge to obtain the global joint debugging cost (mixed loss function), and the target joint debugging cost and the global joint debugging cost are subjected to knowledge collection to obtain the target debugging cost data corresponding to the consultation response reference text record.

The thought of determining the target joint debugging cost can refer to the determination thought of the initial joint debugging cost according to the first consultation response frequent knowledge. According to the second frequent knowledge of the consultation response, the thinking of determining the global joint debugging cost can be in various modes, for example, any consultation response reference text can be extracted from the global consultation response reference text to serve as the current consultation response reference text for determining the joint debugging cost, the joint debugging cost of the global consultation response reference text is determined according to the current consultation response reference text and the second frequent knowledge of the consultation response, the basic global joint debugging cost is obtained, and the step of extracting any consultation response reference text from the global consultation response reference text to serve as the current consultation response reference text for determining the joint debugging cost is carried out until all consultation response reference texts in the global consultation response reference text serve as the current consultation response reference text to determine the joint debugging cost, so that the basic global joint debugging cost corresponding to each consultation response reference text in the global consultation response reference text is obtained, and knowledge is collected to obtain the global joint debugging cost.

Further, according to the current consultation response reference text and the second consultation response frequent knowledge, calculating the joint debugging cost of the global consultation response reference text, and the thought of obtaining the basic global joint debugging cost can refer to the initial joint debugging cost or the determination thought of the target joint debugging cost.

In addition, knowledge collection is performed on the target joint debugging cost and the global joint debugging cost, multiple ideas can be used for obtaining target debugging cost data corresponding to the consultation response reference text record, for example, weight factors of the target joint debugging cost and the global joint debugging cost can be obtained, the target joint debugging cost and the global joint debugging cost are weighted according to the weight factors, knowledge collection is performed on the weighted target joint debugging cost and the global joint debugging cost, and target debugging cost data (loss information) corresponding to the consultation response reference text record can be obtained.

S104, carrying out knowledge collection on the target consultation response frequent item knowledge and the adjusted consultation response frequent item knowledge to obtain linkage consultation response frequent item knowledge for inquiring the consultation response text.

The linked frequent knowledge of the consultation response may be all frequent knowledge of the consultation response for the consultation response text query.

The concept of knowledge collection of the target frequent knowledge and the adjusted frequent knowledge may be various, and examples may include the following related content.

For example, the preset frequent consultation response knowledge set may be adjusted based on the adjusted frequent consultation response knowledge to obtain an adjusted frequent consultation response knowledge set, and the target frequent consultation response knowledge is added to the adjusted frequent consultation response knowledge set to obtain the linked frequent consultation response knowledge, or the target frequent consultation response knowledge and the adjusted frequent consultation response knowledge may be directly combined to obtain the linked frequent consultation response knowledge.

S105, responding to the consultation response text query application, carrying out consultation response vector mining on the large data of the consultation response text to be queried contained in the consultation response text query application, and querying final large data of the consultation response text in the large data of the initial consultation response text and the large data of the target consultation response text according to the determined knowledge of the frequent items of the consultation response to be queried and the knowledge of the linked frequent items of the consultation response.

The concept of performing the query response vector mining on the large data of the query response text to be queried included in the query response text query application may be various, for example, a first query response text query layer in a debugged query response text query algorithm may be used to perform the query response vector mining on the large data of the query response text to be queried to obtain knowledge of frequent items of the query response text to be queried of the large data of the query response text to be queried, where the first query response text query layer in the debugged query response text query algorithm may be a first query response text query layer obtained after collaborative debugging on the debugged first query response text query layer.

After the knowledge of the frequent consultation response item to be queried is determined, final large consultation response text data can be determined in the large initial consultation response text data and the large target consultation response text data according to the knowledge of the frequent consultation response item to be queried and the knowledge of the frequent consultation response item to be queried, for example, a knowledge vector difference value between the knowledge of the frequent consultation response item to be queried and each knowledge of the frequent consultation response item to be queried can be calculated, a first knowledge vector difference value is obtained, the first knowledge vector difference value is organized, final large consultation response text data is determined in the large initial consultation response text data and the large target consultation response text data according to an organization result, or a clustering center corresponding to the large initial consultation response text data can be adjusted according to the knowledge of the frequent consultation response item to be queried, a target catalog is extracted from the adjusted catalog corresponding to the cluster center, and final large consultation response text data is determined in the large initial consultation response text data and the large target consultation response text data according to an organization result.

The concept of determining the final large data of the consultation response text in the initial large data of the consultation response text and the target large data of the consultation response text according to the sorting result (sorting result) of the first knowledge vector difference value may be various, for example, the first large data of the consultation response text with the number K1 before the sorting result may be extracted from the initial large data of the consultation response text and the target large data of the consultation response text as the final large data of the consultation response text, or the first large data of the consultation response text with the minimum first knowledge vector difference value may be extracted from the initial large data of the consultation response text and the target large data of the consultation response text as the final large data of the consultation response text, or the first large data of the candidate large data of the consultation response text is compared with the preset difference threshold, the first large data of the first knowledge vector difference value of the candidate large data of the consultation response text is extracted from the candidate large data of the consultation response text, at least one group of the target large data of the consultation text is obtained, the target large data of the consultation response text is classified according to the first large data of the number of the target large data of the consultation text, the target large data of the consultation text is extracted from the target large data of the consultation text is compared with the preset difference threshold, and extracting the large data of the consultation response text with the preset number K2 with the minimum knowledge vector difference value from the large data of the consultation response text of the target type as the large data of the final consultation response text.

And extracting at least one group of candidate query text big data corresponding to the target catalogue from the initial query response text big data and the target query response text big data according to the target catalogue, calculating a knowledge vector difference value between the to-be-queried query response frequent item knowledge and the query response frequent item knowledge of the candidate query text big data to obtain a second knowledge vector difference value, and extracting final query response text big data from the candidate query text big data according to the second knowledge vector difference value.

The query mode of adopting the catalogue corresponding to the clustering center can be cluster-based distributed query, such as extracting initial consultation response frequent knowledge of initial consultation response text big data, clustering the initial consultation response frequent knowledge to obtain at least one clustering center, taking the clustering center as the queried catalogue, and establishing the relativity between the catalogue and the initial consultation response text big data. At this time, responding to the consultation response text query application, extracting the large data of the consultation response text to be queried contained in the consultation response text query application, finding the nearest catalogue according to the knowledge of the frequent items of the consultation response text to be queried of the large data of the consultation response text to be queried, obtaining the associated large data of the consultation response text of the catalogues to obtain candidate consultation response text query recall, calculating the knowledge vector difference value between the knowledge of the frequent items of the consultation response text and the knowledge of the frequent items of the consultation response text to be queried, and sorting from small to large, wherein the first K1 pieces of large data of the initial consultation response text in the sorting are taken as the output query text. When the large data of the initial consultation response text has the supplementary large data of the target consultation response text, the large data of the target consultation response text is subjected to consultation response vector mining, the knowledge of the frequent items of the initial consultation response of the large data of the initial consultation response text is adjusted, the knowledge of the frequent items of the target consultation response and the knowledge of the frequent items of the adjusted consultation response are combined to obtain the knowledge of the frequent items of the linkage consultation response, and then the cluster center and the corresponding catalogues of the cluster center are updated according to the complete characteristics, so that the optimized inquiry can be realized under the partially adjusted characteristics.

Optionally, after determining the final big data of the consultation response text corresponding to the big data of the consultation response text to be queried, the big data of the consultation response text to be queried can be distinguished or identified according to the category of the final big data of the consultation response text. The thinking of distinguishing or identifying may be various, for example, when the number of the final consultation answer text big data is 1, the distinguishing or identifying result of the consultation answer text big data to be inquired may be the category of the final consultation answer text big data, when the number of the final consultation answer text big data is multiple, the final consultation answer text big data may be distinguished, then the text big data number of the final consultation answer text big data of each category is obtained, the target consultation answer text big data category with the largest text big data number is extracted from the text categories, the target consultation answer text big data category is used as the distinguishing or identifying result of the consultation answer text big data to be inquired, or when the number of the final consultation answer text big data is multiple, the final consultation answer text big data is distinguished, then the sum of knowledge vector difference values corresponding to each text category is determined, and the text category with the minimum sum of knowledge vector difference values is used as the distinguishing or identifying result of the consultation answer text big data to be inquired.

After initial consultation response text big data, initial consultation response frequent item knowledge corresponding to the initial consultation response text big data and at least one group of supplementary target consultation response text big data are obtained, consultation response vector mining is carried out on the target consultation response text big data to obtain target consultation response frequent item knowledge corresponding to the target consultation response text big data, then the initial consultation response frequent item knowledge is adjusted according to the target consultation response frequent item knowledge to obtain adjusted consultation response frequent item knowledge, then the target consultation response frequent item knowledge and the adjusted consultation response frequent item knowledge are subjected to knowledge collection to obtain linkage consultation response frequent item knowledge for consultation response text query, consultation response vector mining is carried out on to-be-queried consultation response text big data contained in the consultation response text query application, and final consultation response text big data are determined in the initial consultation response text big data and the target response text big data according to the determined to-be-queried consultation response frequent item knowledge and the linkage consultation response frequent item knowledge; in view of the fact that the embodiment of the invention can transform the initial consultation response frequent knowledge into the knowledge network of the target consultation response frequent knowledge, the consultation response frequent knowledge for inquiring the consultation response text can be timely adjusted, the initial consultation response text big data does not need to be subjected to consultation response vector mining again, the operation load required by the consultation response vector mining is effectively saved, and the mining timeliness of the consultation response frequent knowledge can be improved, so that the timeliness of inquiring the consultation response text is ensured.

Under some design ideas which can be implemented independently, after the target digital cloud service is optimally updated for the software service based on the software service update requirement and the preset resource use constraint condition, the method can further comprise the technical scheme described in the step four.

Step four, acquiring a user operation data set uploaded by the target service platform system; performing abnormal behavior recognition based on the user operation data set to obtain an abnormal behavior recognition result; issuing the abnormal behavior identification result to the target service platform system; the user operation data set is obtained by data acquisition based on the target digital cloud service after optimization and updating.

It can be understood that the abnormal behavior recognition aiming at the user operation data set is performed on the big data optimization server side instead of the target service platform system side, so that on one hand, the operation pressure of the target service platform system can be reduced, and on the other hand, the accurate and timely abnormal behavior recognition can be performed by means of the big data optimization server with stronger calculation power, so that risk prompt is performed on the target service platform system in time, and the data security of the target service platform system in operation is ensured.

Under some design ideas which can be implemented independently, the abnormal behavior recognition is performed based on the user operation data set, and the abnormal behavior recognition result is obtained, which may include the following contents: extracting a first page behavior description field and a second page behavior description field of the user operation data set; the first page behavior description field indicates a page behavior description field with the behavior track update frequency in the user operation data set lower than or equal to a first determination value, and the second page behavior description field indicates a page behavior description field with the behavior track update frequency in the user operation data set higher than the first determination value; performing disturbance cleaning treatment on the second page behavior description field to obtain a third page behavior description field; extracting a page behavior description field from a first reference operation data set generated based on the first page behavior description field to obtain a behavior description field extraction result; the behavior description field extraction result comprises a fourth page behavior description field, wherein the fourth page behavior description field indicates the page behavior description field with the behavior track updating frequency higher than a second determination value in the first reference operation data set; based on the fourth page behavior description field, fusing the third page behavior description field and the first reference operation data set to obtain a target user operation data set, wherein the target user operation data set represents a disturbance cleaning result of the user operation data set; and carrying out abnormal behavior recognition on the target user operation data set based on the convolutional neural network which is trained in advance, and obtaining an abnormal behavior recognition result. Wherein the decision value can be understood as a threshold value.

It can be understood that, before abnormal behavior identification is performed, the user operation data set can be subjected to disturbance cleaning based on the page behavior description field and the behavior track update frequency, so that normal operation behaviors can be primarily filtered out, a more simplified target user operation data set is obtained, and an abnormal behavior identification result corresponding to the target user operation data set is efficiently and accurately mined through a convolutional neural network which is trained in advance.

In the embodiment of the invention, the convolutional neural network which is trained in advance can be used for determining the decision score corresponding to the behavior preference vector by mining the behavior preference vector of the target user operation data set and then carrying out decision analysis, and then obtaining the abnormal behavior recognition result through the decision score. For example, if the decision score is greater than the set score, it may be determined that the abnormal behavior recognition result corresponding to the target user operation data set is "abnormal", otherwise it may be determined that it is "normal". Thus, the risk prompt can be carried out for the target service platform system when the abnormal behavior recognition result is abnormal.

Under some design ideas that can be implemented independently, the behavior description field extraction result further includes a fifth page behavior description field, where the fifth page behavior description field indicates a page behavior description field with a behavior trace update frequency lower than or equal to the second determination value in the first reference operation data set, the fourth page behavior description field and the fifth page behavior description field are obtained based on feature encoding, and based on the fourth page behavior description field, fusing the third page behavior description field and the first reference operation data set to obtain a target user operation data set, where the method includes: determining a first degree of correlation between the fourth page behavior description field and the third page behavior description field; and performing feature decoding on the third page behavior description field, the fourth page behavior description field and the fifth page behavior description field based on the first correlation degree and the second determination value to obtain the target user operation data set.

When the method is applied to the embodiment of the invention, when the consultation response text query application is received, the final consultation response text big data is determined by utilizing the consultation response characteristic analysis strategy; the final consultation response text big data is consultation response text big data of the target digital cloud service; obtaining software service optimization requirements aiming at the target digital cloud service according to the final consultation response text big data; based on the software service update requirement and a preset resource use constraint condition, optimizing and updating the software service for the target digital cloud service; the resource use constraint condition is used for realizing overload protection of a target service platform system, and the target service platform system is a platform system corresponding to the target digital cloud service.

The foregoing is only a specific embodiment of the present invention. Variations and alternatives will occur to those skilled in the art based on the detailed description provided herein and are intended to be included within the scope of the invention.

Claims

1. A big data optimization method applied to a digital cloud service, characterized in that the method is applied to a big data optimization server, and the method comprises:

2. The method of claim 1, wherein the determining final counsel response text big data using a counsel response feature analysis strategy comprises:

3. The method of claim 2, wherein the adjusting the initial counseling response frequent knowledge using the target counseling response frequent knowledge to obtain adjusted counseling response frequent knowledge comprises:

4. The method as set forth in claim 3, wherein the transforming the initial frequent counseling knowledge to the knowledge network of the target frequent counseling knowledge by the knowledge transforming layer of the debugged frequent counseling text query algorithm to obtain the adjusted frequent counseling knowledge by comparing the results, further includes:

5. The method of claim 4 wherein the obtaining a consultation response reference text record comprises:

6. The method of claim 4, wherein said determining advisory response reference joint text for debugging in the initial and target advisory response reference text sets comprises:

7. The method as set forth in claim 4, wherein the debugging the first consultation response text query layer of the general consultation response text query algorithm using the initial consultation response reference text set and the consultation response reference joint text to obtain a debugged first consultation response text query layer, comprising:

8. The method as set forth in claim 5, wherein the collaborative debugging of the first post-debugging consultation response text query layer and the knowledge transformation layer in the general consultation response text query algorithm based on the consultation response reference text record and the consultation response reference joint text to obtain a debugged consultation response text query algorithm includes: adopting the debugged first consultation response text query layer to carry out consultation response vector mining on the consultation response reference text in the consultation response reference text record, so as to obtain first consultation response frequent knowledge; adopting the second consultation response text query layer to carry out consultation response vector mining on the consultation response reference text in the consultation response reference text record, and adopting a knowledge transformation layer in the general consultation response text query algorithm to transform the mined frequent knowledge to obtain second consultation response frequent knowledge; optimizing the debugged first consultation response text query layer and knowledge transformation layer by utilizing the first consultation response frequent knowledge, the second consultation response frequent knowledge and the consultation response reference joint text to obtain a debugged consultation response text query algorithm;

The optimizing the first consultation response text query layer and the knowledge transformation layer after the debugging by using the first consultation response frequent knowledge, the second consultation response frequent knowledge and the consultation response reference joint text to obtain a debugged consultation response text query algorithm comprises the following steps: extracting a combined sample set formed by the consultation response reference texts in the target consultation response reference text set from the consultation response reference text set to obtain a target consultation response combined text; extracting a combined sample set consisting of the consultation response reference text in the target consultation response reference text set and the consultation response reference text in the initial consultation response reference text set from the consultation response reference combined text to obtain a global consultation response reference text; determining target debugging cost data corresponding to the consultation response reference text record by utilizing the target consultation response combined text, the global consultation response reference text, the first consultation response frequent item knowledge and the second consultation response frequent item knowledge; optimizing the first consultation response text query layer and the knowledge transformation layer after debugging by using the target debugging cost data to obtain a debugged consultation response text query algorithm;

The determining, by using the target consultation response joint text, the global consultation response reference text, the first consultation response frequent knowledge and the second consultation response frequent knowledge, the target debugging cost data corresponding to the consultation response reference text record includes: determining the joint debugging cost of the target consultation response joint text according to the first consultation response frequent knowledge to obtain the target joint debugging cost; based on the second consultation response frequent knowledge, determining the joint debugging cost of the global consultation response reference text to obtain global joint debugging cost; carrying out knowledge collection on the target joint debugging cost and the global joint debugging cost to obtain target debugging cost data corresponding to the consultation response reference text record;

wherein, based on the second frequent knowledge of consultation response, determining the joint debugging cost of the global consultation response reference text to obtain a global joint debugging cost includes: extracting any one consultation response reference text from the global consultation response reference text as a current consultation response reference text for determining joint debugging cost; determining the joint debugging cost of the global consultation response reference text by utilizing the current consultation response reference text and the second consultation response frequent knowledge to obtain basic global joint debugging cost; jumping to the step of extracting any consultation response reference text from the global consultation response reference text as the current consultation response reference text for determining the joint debugging cost until all the consultation response reference texts in the global consultation response reference text are used as the current consultation response reference text for determining the joint debugging cost, and obtaining the basic global joint debugging cost corresponding to each consultation response reference text in the global consultation response reference text; and carrying out knowledge collection on the basic global joint debugging cost to obtain the global joint debugging cost.

9. The method as set forth in claim 2, wherein the determining final counseling response text big data from the initial counseling response text big data and the target counseling response text big data according to the determined counseling response frequent knowledge and the linked counseling response frequent knowledge includes:

10. A big data optimization server, comprising: a memory and a processor; the memory is coupled to the processor; the memory is used for storing computer program codes, and the computer program codes comprise computer instructions; wherein the computer instructions, when executed by the processor, cause the big data optimization server to perform the method of any of claims 1-9.