CN117454146A - Operation and maintenance strategy determining method, device, robot, storage medium and program product - Google Patents

Operation and maintenance strategy determining method, device, robot, storage medium and program product Download PDF

Info

Publication number
CN117454146A
CN117454146A CN202311386702.3A CN202311386702A CN117454146A CN 117454146 A CN117454146 A CN 117454146A CN 202311386702 A CN202311386702 A CN 202311386702A CN 117454146 A CN117454146 A CN 117454146A
Authority
CN
China
Prior art keywords
maintenance
data
strategy
session data
semantic
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202311386702.3A
Other languages
Chinese (zh)
Inventor
刘昕林
邓巍
张健
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Power Supply Bureau Co Ltd
Original Assignee
Shenzhen Power Supply Bureau Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Power Supply Bureau Co Ltd filed Critical Shenzhen Power Supply Bureau Co Ltd
Priority to CN202311386702.3A priority Critical patent/CN117454146A/en
Publication of CN117454146A publication Critical patent/CN117454146A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/251—Fusion techniques of input or preprocessed data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • G06F40/35—Discourse or dialogue representation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Machine Translation (AREA)

Abstract

The application relates to an operation and maintenance strategy determining method, an operation and maintenance strategy determining device, a robot, a storage medium and a program product. The method comprises the following steps: acquiring at least one operation and maintenance session data; extracting features of the operation and maintenance session data to obtain operation and maintenance semantic features; selecting a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance semantic features; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system. According to the technology of the application, the operation and maintenance strategy determination efficiency and the determination result accuracy are improved.

Description

Operation and maintenance strategy determining method, device, robot, storage medium and program product
Technical Field
The present disclosure relates to the field of operation and maintenance technologies, and in particular, to an operation and maintenance policy determining method, an apparatus, a robot, a storage medium, and a program product.
Background
The operation and maintenance are essentially to the network, server and service of the business system, based on some operation and maintenance strategies, at each stage of its life cycle, so as to ensure that the business system reaches a consistent and acceptable state in terms of cost, stability and efficiency.
The conventional technology generally needs to input a large number of professional operation and maintenance personnel, and determines an operation and maintenance strategy manually according to the actual monitoring condition of a service system for subsequent operation and maintenance processing. However, this method requires that the operation and maintenance personnel have high expertise, is limited by enthusiasm, technical ability, subjective consciousness and the like of the operation and maintenance personnel, and has poor operation and maintenance policy determination efficiency and determination result accuracy.
Disclosure of Invention
Based on this, it is necessary to provide an operation and maintenance policy determining method, apparatus, robot, storage medium and program product that are more efficient and have better accuracy, in view of the above-mentioned technical problems.
In a first aspect, the present application provides an operation and maintenance policy determining method, including:
acquiring at least one operation and maintenance session data;
extracting features of the operation and maintenance session data to obtain operation and maintenance semantic features;
selecting a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance semantic features; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system.
In one embodiment, the feature extraction of the operation and maintenance session data to obtain operation and maintenance semantic features includes: determining semantic association conditions among the operation and maintenance-free session data, and extracting semantic association features among different operation and maintenance session data according to the semantic association conditions; and taking the semantic association features as the operation and maintenance semantic features.
In one embodiment, the determining the semantic association condition between the operation and maintenance session data, and extracting the semantic association features between different operation and maintenance session data according to the semantic association condition includes: and determining semantic association conditions among the unnecessary operation and maintenance session data based on the trained language characterization model, and extracting semantic association features among different operation and maintenance session data according to the semantic association conditions.
In one embodiment, the language characterization model is trained in the following manner: acquiring a training session set; wherein the training session set comprises a plurality of training session data; according to different training session data and semantic association labels among different training session data, constructing a plurality of training sample pairs comprising two training session data; and carrying out parameter fine adjustment on the pre-trained language characterization model according to the plurality of training sample pairs and the semantic association condition of each training sample pair.
In one embodiment, the training sample pair comprises a training positive sample pair and a training negative sample pair; correspondingly, according to the semantic association labels between different training session data and different training session data, a plurality of training sample pairs comprising two training session data are constructed, including: according to any two training session data with semantic association, a training positive sample pair is constructed; and constructing a training negative sample pair according to any two training session data without semantic association.
In one embodiment, the performing parameter fine tuning on the pre-trained language characterization model according to the semantic association situations between the plurality of training sample pairs and each training sample pair includes: aiming at any training sample pair, masking word segmentation results with preset proportion in the training sample pair so as to update the training sample pair; inputting the updated training sample pair into a pre-trained language characterization model to obtain sample semantic features; predicting semantic association categories of different training session data in the training sample pair according to the sample semantic features; and carrying out parameter fine adjustment on the pre-trained language characterization model according to the semantic association category and the actual semantic association condition of the training sample pair.
In one embodiment, the selecting, according to the operation and maintenance semantic feature, a target operation and maintenance policy from at least one candidate operation and maintenance policy includes: acquiring a preset first corresponding relation; the first corresponding relation is a corresponding relation between different reference semantic features and different candidate operation and maintenance strategies; and selecting a target operation and maintenance strategy corresponding to the operation and maintenance semantic feature from at least one candidate operation and maintenance strategy according to the first corresponding relation.
In one embodiment, the first correspondence is constructed in the following manner: acquiring a reference session data set comprising a plurality of reference session data, and a candidate policy data set comprising a plurality of candidate policy data; respectively extracting features of each reference session data to obtain corresponding reference semantic features; recall candidate policy data from the candidate policy data set according to the reference semantic features; and establishing a first corresponding relation between the reference semantic features and recalled candidate strategy data corresponding to the reference semantic features.
In one embodiment, recalling candidate policy data from the candidate policy data set according to the reference semantic features comprises: and recalling candidate strategy data from the candidate strategy data set according to the reference semantic features based on a strategy recall model.
In one embodiment, the policy recall model is trained in the following manner: acquiring a sample session set; the sample session set comprises a plurality of sample session data and labeling strategy data corresponding to each sample session data; extracting features of the sample session data to obtain sample session features; inputting the sample session characteristics into a pre-constructed strategy recall model to obtain recall strategy data of corresponding sample session data; and carrying out model training on the strategy recall model according to recall strategy data of each sample session data and corresponding labeling strategy data.
In one embodiment, the model training of the policy recall model according to the recall policy data and the corresponding labeling policy data of each sample session data includes: determining recall accuracy of the sample session set according to the ranking order of recall probabilities corresponding to the labeling strategy data of the sample session data in the candidate strategy data set; determining the policy accuracy of the sample session set according to the consistency of recall policy data of each sample session data and corresponding labeling policy data; and carrying out model training on the strategy recall model according to the recall accuracy and/or the strategy accuracy.
In one embodiment, the determining the recall accuracy of the sample session set according to the ranking order of the recall probabilities corresponding to the labeling policy data of each sample session data in the candidate policy data set includes: counting the maximum recall probability average value of the labeling strategy data with the maximum recall probability corresponding to the labeling strategy data of each sample session data in the candidate strategy data set, and taking the average value determination result as a first recall accuracy; determining a second recall accuracy according to the ranking order of recall probabilities of the labeling strategy data of each sample session data in the candidate strategy data set; generating the recall accuracy including a first recall accuracy and/or a second recall accuracy.
In one embodiment, the acquiring a reference session data set including a plurality of reference session data includes: acquiring an initial session data set comprising a plurality of reference session data; performing cluster analysis on each reference session data in the initial session data set; and respectively selecting different types of reference session data from the initial session data set, and constructing the reference session data set.
In one embodiment, obtaining a candidate policy data set comprising a plurality of candidate policy data comprises: acquiring an initial policy data set comprising a plurality of candidate policy data; performing cluster analysis on each candidate strategy data in the initial strategy data set; and respectively selecting candidate strategy data of different categories from the initial strategy data set, and constructing the candidate strategy data set.
In a second aspect, the present application further provides an operation and maintenance policy determining device, including:
the operation and maintenance session data acquisition module is used for acquiring at least one operation and maintenance session data;
the operation and maintenance semantic feature extraction module is used for extracting features of the operation and maintenance session data to obtain operation and maintenance semantic features;
the target operation and maintenance strategy selection module is used for selecting a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance semantic characteristics; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system.
In a third aspect, the present application also provides a robot, including a memory and a processor, the memory storing a computer program, the processor implementing the following steps when executing the computer program:
acquiring at least one operation and maintenance session data;
extracting features of the operation and maintenance session data to obtain operation and maintenance semantic features;
selecting a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance semantic features; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system.
In a fourth aspect, the present application also provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:
acquiring at least one operation and maintenance session data;
extracting features of the operation and maintenance session data to obtain operation and maintenance semantic features;
selecting a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance semantic features; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system.
In a fifth aspect, the present application also provides a computer program product comprising a computer program which, when executed by a processor, performs the steps of:
Acquiring at least one operation and maintenance session data;
extracting features of the operation and maintenance session data to obtain operation and maintenance semantic features;
selecting a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance semantic features; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system.
According to the operation and maintenance strategy determining method, the device, the robot, the storage medium and the program product, the operation and maintenance semantic features are obtained by extracting the features of at least one operation and maintenance session data, and the target operation and maintenance strategy is automatically selected from at least one candidate operation and maintenance strategy for representing the operation and maintenance processing mode of the system according to the obtained operation and maintenance semantic features, so that the operation and maintenance strategy determining efficiency is improved. Meanwhile, professional operation and maintenance personnel are not required to be introduced in the process to determine the operation and maintenance strategy, so that the labor cost of the operation and maintenance process is reduced, the professional threshold of the operation and maintenance personnel is reduced, and the influence of human factors on the accuracy of the operation and maintenance strategy determination result is avoided. In addition, the operation and maintenance semantic features obtained by feature extraction are based on at least one operation and maintenance session data, and the target operation and maintenance strategy is determined, so that the influence caused by different operation and maintenance session data can be comprehensively considered in the process of determining the target operation and maintenance strategy, the richness of data referenced in the process of determining the operation and maintenance strategy is improved, and the accuracy of the determination result of the operation and maintenance strategy is further improved.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the related art, the drawings that are required to be used in the embodiments or the related technical descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and other drawings may be obtained according to the drawings without inventive effort for a person having ordinary skill in the art.
Fig. 1 is an application environment diagram of an operation and maintenance policy determining method provided in an embodiment of the present application;
fig. 2 is a flow chart of an operation and maintenance policy determining method provided in an embodiment of the present application;
FIG. 3 is a schematic flow chart of a method for determining operation and maintenance semantic features according to an embodiment of the present application;
fig. 4 is a flow chart of a target operation and maintenance policy selection method provided in an embodiment of the present application;
FIG. 5 is a flowchart of another method for determining an operation and maintenance policy according to an embodiment of the present application;
fig. 6 is a block diagram of an operation and maintenance policy determining device according to an embodiment of the present application;
fig. 7 is an internal structural diagram of a robot according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.
The operation and maintenance strategy determination method provided by the embodiment of the application can be applied to an application environment shown in fig. 1. Wherein the terminal 102 communicates with the server 104 via a network. The data storage system may store data that the server 104 needs to process. The data storage system may be integrated on the server 104 or may be located on a cloud or other network server. The terminal 102 obtains at least one operation and maintenance session data; the terminal 102/the server 104 performs feature extraction on the operation and maintenance session data to obtain operation and maintenance semantic features; the terminal 102/server 104 selects a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance semantic features; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system. The terminal 102 may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, internet of things devices and portable wearable devices, where the internet of things devices may be at least one of smart speakers, smart televisions, smart air conditioners, smart robots, smart vehicle devices, and the like. The portable wearable device may be at least one of a smart watch, a smart bracelet, a headset, and the like. The server 104 may be implemented as a stand-alone server or as a server cluster of multiple servers.
In an exemplary embodiment, as shown in fig. 2, an operation and maintenance policy determining method is provided, and the method is applied to the terminal 102 in fig. 1 for illustration. It will be appreciated that the method may also be applied to a server, and may also be applied to a system including a terminal and a server, and implemented through interaction between the terminal and the server, which is not limited in any way in this application.
Referring to the operation and maintenance policy determining method shown in fig. 2, the method includes:
s210, acquiring at least one operation and maintenance session data.
Wherein the operation and maintenance session data is used for characterizing the session data describing the operation and maintenance requirements of the data provider (such as the operation and maintenance demander). The operation and maintenance session data may include session data input by a period to which the current time belongs, where the time length of the period to which the current time belongs may be set or adjusted by a technician according to actual needs or experience, which is not limited in any way in the present application. By way of example, the operation and maintenance session data may include current session data entered at a current time and historical session data corresponding to a historical period (e.g., adjacency) associated with the current time. The specific time length, start-stop time and the like of the history period associated with the current moment can be set or adjusted by a technician according to actual requirements or experience, or can be determined through a large number of experiments, and the method is not limited in any way.
It can be understood that, because the number of the operation and maintenance session data is at least one, the richness of the operation and maintenance requirement information carried by the operation and maintenance session data is improved, and when the operation and maintenance session data is the session data input at different moments, the loss of the relevance between the operation and maintenance requirements at different moments is avoided because the operation and maintenance session data is carried with the operation and maintenance requirements at different moments, and a foundation is laid for the improvement of the accuracy of the subsequent operation and maintenance strategy determination result.
And S220, extracting features of the operation and maintenance session data to obtain operation and maintenance semantic features.
Optionally, the operation and maintenance session data can be directly encoded to obtain operation and maintenance semantic features; or alternatively, the operation and maintenance session data can be subjected to coding processing, and the characteristic extraction is performed on the coding processing result, so that the influence of irrelevant information is eliminated, and the operation and maintenance semantic characteristics are obtained. The encoding process may be implemented by at least one encoding mode in the related art, and the feature extraction may be implemented by at least one feature extraction mode in the related art, which is not limited in this application.
In order to improve the adaptation capability of identifying different types of operation and maintenance sessions in the process of determining a target operation and maintenance strategy, in an optional embodiment, if the operation and maintenance session data are text type data, feature extraction is directly performed on the operation and maintenance session data to obtain operation and maintenance semantic features; and if the operation and maintenance session data are voice type data, converting the operation and maintenance session data into text type data, and extracting features of a conversion result to obtain operation and maintenance semantic features. The method has the advantages that different types of operation and maintenance session data such as text type and voice type can be adapted in the process of determining the target operation and maintenance strategy, and flexibility and universality of the operation and maintenance strategy determining process are improved.
S230, selecting a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance semantic features; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system.
In an alternative embodiment, the operation and maintenance intention category to which the operation and maintenance session data belongs can be determined according to the operation and maintenance semantic features; and selecting a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance intention category.
For example, a pre-trained intent classification model may be used to determine, based on the operation and maintenance semantic features, the operation and maintenance intent class to which the operation and maintenance session data belongs. The model training can be performed on the pre-built intention classification model by adopting operation and maintenance session training features extracted by a large number of operation and maintenance session training data and pre-marked intention categories to which each operation and maintenance session training belongs. The intention classification model may be implemented by at least one neural network model in the related art, and the specific network structure of the intention classification model is not limited in this application. For example, the intent classification model may be implemented based on a BiLSTM (Bi-directional Long Short-Term Memory) network.
For example, at least one cluster analysis mode in the related technology can be adopted to perform cluster analysis on the operation and maintenance semantic features, so as to obtain the operation and maintenance intention category to which the operation and maintenance session data corresponding to each operation and maintenance semantic feature belongs.
Correspondingly, according to the operation and maintenance intention category, selecting a target operation and maintenance policy from at least one candidate operation and maintenance policy, wherein the candidate operation and maintenance policy corresponding to the operation and maintenance intention category can be selected as the target operation and maintenance policy according to the pre-constructed corresponding relation between the disagreement graph category and the candidate operation and maintenance policy.
It can be understood that the target operation and maintenance strategy is selected by introducing the corresponding relation between the disagreement graph category and the candidate operation and maintenance strategy, the intention category of the operation and maintenance strategy selection is targeted, and the accuracy of the operation and maintenance strategy selection result is improved.
According to the method and the device for processing the operation and maintenance session data, the operation and maintenance semantic features are obtained through feature extraction of the at least one operation and maintenance session data, and the target operation and maintenance policy is automatically selected from at least one candidate operation and maintenance policy for representing an operation and maintenance processing mode of the system according to the obtained operation and maintenance semantic features, so that the determination efficiency of the operation and maintenance policy is improved. Meanwhile, professional operation and maintenance personnel are not required to be introduced in the process to determine the operation and maintenance strategy, so that the labor cost of the operation and maintenance process is reduced, the professional threshold of the operation and maintenance personnel is reduced, and the influence of human factors on the accuracy of the operation and maintenance strategy determination result is avoided. In addition, the operation and maintenance semantic features obtained by feature extraction are based on at least one operation and maintenance session data, and the target operation and maintenance strategy is determined, so that the influence caused by different operation and maintenance session data can be comprehensively considered in the process of determining the target operation and maintenance strategy, the richness of data referenced in the process of determining the operation and maintenance strategy is improved, and the accuracy of the determination result of the operation and maintenance strategy is further improved.
Based on the technical solutions of the foregoing embodiments, the present application further provides an optional embodiment, where the determining process of the operation and maintenance semantic features of S220 is described in detail. It should be noted that, in the embodiments of the present application, parts not described in detail may be referred to the related expressions of other embodiments, which are not described herein.
Referring to fig. 3, a method for determining operation and maintenance semantic features includes:
s310, determining semantic association conditions among the operation and maintenance-free session data, and extracting semantic association features among different operation and maintenance session data according to the semantic association conditions.
S320, taking the semantic association features as operation and maintenance semantic features.
The semantic association condition is used for representing whether semantic association exists between different operation and maintenance session data; the semantic association features are used to structurally characterize semantic associations between different operation and maintenance session data.
In an alternative embodiment, whether semantic association exists between the operation and maintenance session data can be determined by determining the operation and maintenance intention categories to which the different operation and maintenance session data belong and according to the association condition between the operation and maintenance intention categories; semantic association features between the operation and maintenance session data with the semantic association are extracted.
The determination manners of the operation and maintenance intention types to which the different operation and maintenance session data belong can refer to the related expressions of the foregoing embodiments, and are not described herein.
Optionally, extracting semantic association features between the operation and maintenance session data with semantic association may be: respectively extracting initial semantic features of different operation and maintenance session data; and carrying out feature fusion on the initial semantic features among different operation and maintenance session data to obtain semantic association features. The feature fusion can be realized by at least one mode of splicing fusion, weighting fusion and the like, and the specific adopted feature fusion mode is not limited in the application.
Or alternatively, extracting semantic association features between the operation and maintenance session data with semantic association can be: carrying out data fusion on operation and maintenance session data with operation and maintenance association; and extracting features of the data fusion result to obtain semantic association features. The data fusion can be realized by at least one mode of splicing fusion, weighting fusion and the like, and the specific adopted characteristic fusion mode is not limited in the application.
In another alternative embodiment, the semantic association condition between the different operation and maintenance session data can be determined based on the trained language characterization model, and the semantic association features between the different operation and maintenance session data can be extracted according to the semantic association condition. The language characterization model can be implemented by adopting at least one neural network model in the related technology, and the specific network structure of the language characterization model is not limited in the application. In one particular implementation, the language characterization model may be a BERT (Bidirectional Encoder Representation from Transformers, converter-based bi-directional coded representation) model.
For example, the language characterization model may be first pre-trained, and then parameter fine-tuned based on downstream tasks. It can be appreciated that the goal of pre-training is to train on unlabeled data first, to initialize the network parameters of the language characterization model; the purpose of fine tuning is to carry out parameter adjustment again on the pre-trained language characterization model through the marking data of the downstream task (corresponding to the operation and maintenance task of the application), so that the language characterization model can adapt to the specific application scene of the downstream task.
Alternatively, the language characterization model may be trained in the following manner: acquiring a training session set; wherein the training session set comprises a plurality of training session data; according to different training session data and semantic association labels among different training session data, constructing a plurality of training sample pairs comprising two training session data; and carrying out parameter fine adjustment on the pre-trained language characterization model according to the plurality of training sample pairs and the semantic association condition of each training sample pair.
The training session data may be understood as operation and maintenance session data of a training operation and maintenance demander serving as a training sample; the semantic association tag between different training session data is used to characterize whether there is a semantic association between different training session data, which may be determined by manual labeling or other methods, and this application is not limited in any way. The pre-training mode of the language characterization model can be implemented by at least one of related technologies, which is not limited in the application.
For example, the training sample pair may include a training positive sample pair characterizing a positive sample and/or a training negative sample pair characterizing a negative sample. Correspondingly, constructing a plurality of training sample pairs including two training session data according to different training session data and semantic association labels between different training session data may include: constructing a training positive sample pair (for example, can be marked as IsNext) according to any two training session data with semantic association; from any two training session data for which there is no semantic association, a training negative sample pair (e.g., which may be labeled NotNext) is constructed. The method has the advantages that the diversity of training sample pairs can be improved, the adaptability of the operation and maintenance scene of the language characterization model after fine adjustment is improved, the situation of scene mismatch is avoided, and the accuracy of the extracted semantic association features is improved. The specific number and numerical ratio of the training positive sample pair and the training negative sample pair can be set or adjusted by a technician according to experience or actual conditions, or can be determined through a plurality of experiments, and the application is not limited in any way.
In one specific implementation, statement a and statement B may be combined into a training sample pair; if the sentence B is the actual next sentence following the sentence A, the training sample pair constructed by the sentence A and the sentence B is a training positive sample pair; if the sentence B is not the actual next sentence following the sentence a, or if the sentence B is a random sentence in the corpus, the training sample pair constructed by the sentence a and the sentence B is a training negative sample pair.
Optionally, in order to facilitate subsequent model training, a class identifier (CLS) may be added before the previous training session data and an interval identifier (SEP) may be added between the previous training session data and the subsequent training session data in the process of constructing the training sample pair, so as to construct an NSP (Next Sentence Prediction ) task, so that the language characterization model trained subsequently can understand the relationship between different operation and maintenance session sentences. It is noted that when one operation and maintenance session statement is the next statement of another operation and maintenance session statement, it is indicated that the two have semantic association.
In an alternative embodiment, parameter fine tuning is performed on the pre-trained language characterization model according to a plurality of training sample pairs and semantic association conditions of each training sample pair, including: aiming at any training sample pair, masking word segmentation results with preset proportion in the training sample pair so as to update the training sample pair; inputting the updated training sample pair into a pre-trained language characterization model to obtain sample semantic features; predicting semantic association categories of different training session data in the training sample pair according to the sample semantic features; and carrying out parameter fine adjustment on the pre-trained language characterization model according to the semantic association category and the actual semantic association condition of the training sample pair. Alternatively, the language characterization model may be implemented based on an MLM (Masked Language Model, mask language model) model.
The training sample pair corresponding sample semantic features carry information with semantic association of two training session data contained in the training sample pair. Wherein, the semantic association category can be an association category or a non-association category; the actual semantic association condition is a pre-labeled association category or non-association category. The association category is used for representing that two training session data in the predicted training sample pair have an upper sentence relationship and a lower sentence relationship (namely have semantic association); the non-associated class is used for representing that the two training session data in the predicted training sample pair have no sentence relation (i.e. have no semantic association).
The preset proportion of the word segmentation result of the masking process may be set or adjusted by a technician according to the requirement or an empirical value, or may be repeatedly determined through a plurality of experiments, which is not limited in this application. In one particular implementation, masking tools may be employed to mask 15% of the word segmentation results. It can be understood that through the covering process, each word segmentation result in the training session data can be focused in the model training process, so that the model gradually has more accurate semantic feature extraction capability.
For example, if the training sample pair is the NSP task, the semantic features of the sample may be a category vector corresponding to the category identifier and/or a feature fusion result of the word segmentation vectors corresponding to different word segmentation results output by the language characterization model.
For example, the sample semantic features can be input into a classification network constructed based on the full connection layer and the activation layer, and the semantic association class of the training sample pair corresponding to the sample semantic features is determined according to the class prediction probability output by the classification network. The size of the full-connection layer can be 1024 dimensions, and the input sample semantic features can be one-dimensional vectors with the length of 768; the dimension of the outputted processing result can be 2, and after activation processing is performed through a preset activation function (such as Softmax) of the activation layer, category prediction probability is obtained and is used for representing the semantic association category to which the training sample pair corresponding to the semantic feature of the sample belongs.
It can be understood that by introducing training samples to conduct parameter fine adjustment on the predicted semantic association category and the actual semantic association condition, the pre-trained language characterization model is enabled to have the feature extraction capability of the associated semantics, and the accuracy of the extracted features of the language characterization model is improved.
In a specific implementation manner, word segmentation processing can be performed on different operation and maintenance session data through a preset word segmentation technology, so as to obtain at least one word segmentation result; encoding each word segmentation result to obtain word vectors; extracting features of the word vectors based on the BERT model to obtain semantic association features, and determining importance degree data of each word segmentation result in the feature extraction process; and weighting word segmentation semantic features of different word segmentation results contained in the semantic association features according to the importance degree data to obtain final semantic association features. The preset word segmentation technology and the coding processing mode can be realized by at least one of related technologies, and the application is not limited in any way. For example, the preset word segmentation technique may be Jieba (barking) word segmentation technique, and the encoding processing mode may be single-hot encoding.
The importance degree data is used for representing importance degrees of different word segmentation results, and an IF-IDF (Term Frequency-inverse document Frequency) value can be adopted for numerical quantization.
For example, feature extraction may be performed on the word vector based on the BERT model, to obtain semantic association features, which may include: extracting features of each word vector of the operation and maintenance session data through convertors of different levels in the BERT model to obtain representation (token) vectors of different levels; and synthesizing and splicing the expression vectors of different levels to obtain semantic association features, so that the semantic association features carry semantic information of each word segmentation result under different levels.
It can be understood that the scheme carries out final determination of the semantic association features by introducing the importance degree data of each word segmentation result, so that the semantic association features can have word segmentation pertinence, the feature enhancement of the semantic related word segmentation result and the feature weakening of the semantic independent word segmentation result are realized, and the accuracy and the integrity of the finally determined semantic association features are further improved.
According to the technical scheme, the trained language characterization model is introduced to extract the semantic association features, and the feature extraction of different operation and maintenance session data can be comprehensively realized by the language characterization model, so that specific data processing logic is not required to be concerned, and the method is more universal. The language characterization model can multiplex the pre-trained existing model, and is obtained by fine adjustment of a large amount of training session data on the pre-trained model, so that the trained language characterization model has stronger feature extraction capability and has more operation and maintenance scene pertinence, and the accuracy of semantic association feature extraction results is improved.
According to the method and the device, the semantic association conditions among different operation and maintenance session data are determined, the semantic association features among the different operation and maintenance session data are extracted according to the semantic association conditions, and the semantic association features are used as operation and maintenance semantic features. Because the semantic association condition among different operation and maintenance session data is fully considered in the feature extraction process, the extracted semantic association features can be fused with the associated semantic information of the operation and maintenance session data with semantic association, so that the occurrence of the condition of incomplete information caused by single operation and maintenance session data is avoided, and the richness and the comprehensiveness of the extraction result of the semantic association features are improved.
Based on the above technical solutions, the present application further provides an optional embodiment, in which the target operation and maintenance policy selection process of S230 is described in detail. It should be noted that, for the parts not described in detail in the embodiments of the present application, reference may be made to related expressions of other embodiments, which are not described herein.
Referring to fig. 4, a method for selecting a target operation and maintenance policy includes:
s410, acquiring a preset first corresponding relation; the first corresponding relation is a corresponding relation between different reference semantic features and different candidate operation and maintenance strategies.
S420, selecting a target operation and maintenance strategy corresponding to the operation and maintenance semantic feature from at least one candidate operation and maintenance strategy according to the first corresponding relation.
For example, a similarity between the reference semantic features and the operation semantic features may be determined; and taking the candidate operation and maintenance strategy corresponding to the operation and maintenance semantic features with the similarity larger than the preset similarity threshold as a target operation and maintenance strategy.
The similarity determining manner may be implemented by at least one of related technologies, which is not limited in this application. The preset similarity threshold can be set or adjusted by a technician according to experience or actual requirements, or can be determined through a large number of experiments.
In an alternative embodiment, the first correspondence may be artificially constructed or implemented using at least one of the related techniques.
In order to improve the rationality and effectiveness of the first correspondence, in another alternative embodiment, the construction of the first correspondence may be performed in the following manner: acquiring a reference session data set comprising a plurality of reference session data, and a candidate policy data set comprising a plurality of candidate policy data; respectively extracting features of each reference session data to obtain corresponding reference semantic features; recall candidate policy data from the candidate policy data set according to the reference semantic features; and establishing a first corresponding relation between the reference semantic features and the recalled candidate strategy data corresponding to the reference semantic features.
The reference session data may be understood as known operation and maintenance session data adopted when the first correspondence is constructed, and is not operation and maintenance session data obtained by performing operation and maintenance policy determination this time. The reference semantic feature may be understood as a data feature related to an operation and maintenance requirement carried in the reference session data obtained by extracting a feature from the reference session data.
In one particular implementation, the reference session data may be historical statistical operation and maintenance query statements; the candidate policy data may be operation and maintenance measure data. Different reference session data may correspond to the same or different candidate policy data.
Illustratively, acquiring a reference session data set including a plurality of reference session data may include: acquiring an initial session data set comprising a plurality of reference session data; performing cluster analysis on each reference session data in the initial session data set; and respectively selecting different types of reference session data from the initial session data set to construct a reference session data set. The cluster analysis may be implemented by at least one clustering method in the related art, which is not limited in this application.
Illustratively, obtaining a candidate policy data set including a plurality of candidate policy data may include: acquiring an initial policy data set comprising a plurality of candidate policy data; performing cluster analysis on each candidate strategy data in the initial strategy data set; and respectively selecting candidate strategy data of different categories from the initial strategy data set, and constructing a candidate strategy data set. The cluster analysis may be implemented by at least one clustering method in the related art, which is not limited in this application.
In the process of acquiring the reference session data set and/or the candidate strategy data set, a cluster analysis mode is introduced, so that the merging of the set elements with the same intention category in the corresponding data set is realized, the increase of the data volume caused by the expression difference of the set elements with the same category is avoided, and the data storage and the data transmission data volume of the data set are reduced.
In an alternative embodiment, for any reference session data, if the reference session data is text type data, the feature extraction is directly performed on the reference session data to obtain corresponding reference semantic features; if the reference session data is voice type data, converting the reference session data into text type data, and extracting features of a conversion result to obtain corresponding reference semantic features. The method has the advantages that different types of reference session data can be adapted in the process of collecting the reference session data set, and the convenience and flexibility of acquiring the reference session data set are improved, so that the convenience of the first corresponding relation construction process is improved.
For example, for any reference session data, a pre-trained feature extraction network may be used to perform feature extraction on the reference session data to obtain reference semantic features for establishing the first correspondence. The feature extraction network can be obtained by training with other networks in a class prediction model. The category prediction network comprises a feature extraction network, a feature mapping network and a classification network which are connected in sequence. Optionally, in the training process, the feature extraction network may be used to perform feature extraction on the training session data; mapping the feature extraction result to an intention category space by adopting a feature mapping network; classifying the intention mapping result by adopting a classification network to obtain a predicted intention category; according to the difference condition between the preset intention category of training session data and the corresponding pre-labeled standard intention category, performing iterative training on the feature extraction network, the feature mapping network and the classification network; until the iteration termination condition is satisfied. The iteration termination condition may be that a preset iteration number is met or the model overall tends to converge, which is not limited in the application.
Accordingly, the foregoing feature extraction of the operation and maintenance session data may be implemented by using the feature extraction network obtained by the foregoing training method, which is not limited in this application.
In an alternative embodiment, recall candidate policy data from the candidate policy data set based on the reference semantic features may include: based on the policy recall model, recall candidate policy data from the candidate policy data set according to the reference semantic features. The policy recall model can be constructed by adopting at least one neural network model in the related technology, and the specific network structure of the policy recall model is not limited in the application.
It can be understood that the recall of the candidate strategy data is carried out from the candidate strategy data set by introducing the trained strategy recall model, a specific processing mechanism of the strategy recall model is not required to be concerned, convenience of the recall of the candidate strategy data is improved, multiplexing of the strategy recall model is facilitated, and universality of the strategy recall model is improved.
Illustratively, the policy recall model is trained in the following manner: acquiring a sample session set; the sample session set comprises a plurality of sample session data and labeling strategy data corresponding to each sample session data; extracting features of the sample session data to obtain sample session features; inputting the sample session characteristics into a pre-constructed strategy recall model to obtain recall strategy data of corresponding sample session data; and carrying out model training on the strategy recall model according to recall strategy data of each sample session data and corresponding labeling strategy data.
The sample session data may be understood as operation and maintenance session data adopted when the policy recall model is trained, but not reference session data adopted when the first correspondence is constructed, and also not operation and maintenance session data obtained when the operation and maintenance policy is determined this time. The labeling strategy data can be obtained by adopting a manual labeling or other labeling modes, and the application is not limited in any way. Recall policy data may be understood as candidate policy data recalled from a large number of candidate policy data that has some association with the incoming sample session data. Wherein the plurality of candidate policy data may be the same as or at least partially different from the candidate policy data in the first correspondence construction process, which is not limited in any way in the present application.
In an alternative embodiment, according to recall policy data and corresponding labeling policy data of each sample session data, model training of the policy recall model may include: and carrying out iterative training on the strategy recall model according to the difference condition between the recall strategy data of each sample session data and the corresponding labeling strategy data until the iteration termination condition is met. The iteration termination condition may be that the preset iteration times are met or the model overall tends to converge, and the like, which is not limited in the application. The difference condition can be based on a preset loss function, and numerical quantization is carried out on the determined loss data according to recall strategy data of each sample session data and corresponding labeling strategy data.
In another alternative embodiment, model training of the policy recall model according to recall policy data and corresponding labeling policy data for each sample session data may include: determining recall accuracy of the sample session set according to the ranking order of recall probabilities corresponding to the labeling strategy data of each sample session data in the candidate strategy data set; determining the strategy accuracy of the sample session set according to the consistency of recall strategy data of each sample session data and corresponding labeling strategy data; and performing model training on the strategy recall model according to the recall accuracy and/or the strategy accuracy.
The recall accuracy is used for representing the accuracy of strategy recall behavior; the policy accuracy is used to characterize the accuracy of the policy recall result.
The method comprises the steps that an average value of maximum recall probability of the labeling strategy data with the maximum recall probability corresponding to the labeling strategy data of each sample session data in a candidate strategy data set can be counted, and a mean value determination result is used as a first recall accuracy; determining a second recall accuracy according to the ranking order of recall probabilities of the labeling strategy data of each sample session data in the candidate strategy data set; a recall accuracy rate is generated that includes the first recall accuracy rate and/or the second recall accuracy rate.
In one specific implementation, the first recall accuracy may be determined using the following formula:
wherein |q| is the amount of data in the sample session dataset; c (C) i Recall policy data recalled for the ith sample session data in the sample session set; a is that i Marking strategy data corresponding to the ith sample session data in the sample session set; sigma () is a recall probability mean value determination function; acc@1 is the first recall accuracy.
In another specific implementation, the second recall accuracy may be determined using the following formula:
wherein |q| is the amount of data in the sample session dataset; rank (rank) i Marking strategy data of the ith sample session data in the sample session set, and ranking the recall probability from high to low in the candidate strategy data set; MRR is the second recall accuracy.
According to the technical scheme for generating the recall accuracy by introducing the first recall accuracy and/or the second recall accuracy, the richness of the data referred in the strategy recall model training process is improved, and the effectiveness and the accuracy of recall behaviors of the strategy recall model are improved.
Illustratively, determining the policy accuracy of the sample session set according to the consistency of recall policy data and corresponding labeling policy data of each sample session data may include: determining the sample accuracy according to the consistency of recall strategy data and labeling strategy data of any sample session data aiming at a candidate strategy data set of the sample session data; determining the accuracy of a demand party according to the number of sample session data corresponding to operation and maintenance demand parties to which different sample session data belong and the sample accuracy of the associated sample session data; and determining the policy accuracy according to the number of the operation and maintenance requesters in the sample session set and the accuracy of the requesters.
In one specific implementation, the following formula may be used to determine the sample accuracy:
TP is the number of the collection of the recall strategy data, which is located in the labeling strategy data; FP is the number of recall policy data that is not located in the set to which the annotation policy data pertains.
In another specific implementation, the following formula may be used to determine the demand side accuracy:
wherein, AP i The accuracy of the demand party is the i-th operation and maintenance demand party; p (P) j Sample accuracy of the jth sample session data for the ith requestor; k is the total number of sample session data associated with the ith operation and maintenance requirement party.
In yet another specific implementation, the policy accuracy may be determined using the following formula:
wherein, AP i The accuracy of the demand party is the i-th operation and maintenance demand party; n is the total number of operation and maintenance requesters in the sample session set.
According to the technical scheme, the model training of the strategy recall model is performed by introducing the recall accuracy and/or the strategy accuracy, so that the training of the strategy recall model can be performed under different dimensionalities, and meanwhile, the richness and diversity of the strategy recall model training process are improved.
According to the method and the device for determining the target operation and maintenance strategy, the first correspondence between different reference semantic features and different candidate operation and maintenance strategies is introduced, so that the complex processing of the operation and maintenance semantic features is replaced in a mode of searching and matching through the first correspondence, and the convenience of the determination process of the target operation and maintenance strategy is improved. Meanwhile, in the different operation and maintenance strategy determining processes, the first corresponding relation can be shared, so that parallel processing of data is realized, and the data processing efficiency of a large number of strategy determining processes is improved.
Based on the above technical solutions, the present application further provides an optional embodiment for implementing the operation and maintenance policy determining method. It should be noted that, in the embodiments of the present application, parts not described in detail may be referred to the related expressions of other embodiments, which are not described herein.
Referring to fig. 5, a method for determining an operation and maintenance policy includes:
s510, acquiring at least one operation and maintenance session data.
S520, determining semantic association conditions among the operation and maintenance-free session data based on the trained language characterization model, and extracting semantic association features among different operation and maintenance session data according to the semantic association conditions.
S530, taking the semantic association features as operation and maintenance semantic features.
S540, acquiring a preset first corresponding relation; the first corresponding relation is a corresponding relation between different reference semantic features and different candidate operation and maintenance strategies.
S550, selecting a target operation and maintenance strategy corresponding to the operation and maintenance semantic features from at least one candidate operation and maintenance strategy according to the first corresponding relation; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system.
In an alternative embodiment, the language characterization model is trained in the following manner: acquiring a training session set; wherein the training session set comprises a plurality of training session data; according to any two training session data with semantic association, a training positive sample pair is constructed; according to any two training session data without semantic association, a training negative sample pair is constructed; generating a training sample pair comprising a training positive sample pair and/or a training negative sample pair; aiming at any training sample pair, masking word segmentation results with preset proportion in the training sample pair so as to update the training sample pair; inputting the updated training sample pair into a pre-trained language characterization model to obtain sample semantic features; predicting semantic association categories of different training session data in the training sample pair according to the sample semantic features; and carrying out parameter fine adjustment on the pre-trained language characterization model according to the semantic association category and the actual semantic association condition of the training sample pair.
In an alternative embodiment, the first correspondence is constructed in the following manner: acquiring a reference session data set comprising a plurality of reference session data, and a candidate policy data set comprising a plurality of candidate policy data; respectively extracting features of each reference session data to obtain corresponding reference semantic features; recall candidate policy data from the candidate policy data set according to the reference semantic features based on the policy recall model; and establishing a first corresponding relation between the reference semantic features and the recalled candidate strategy data corresponding to the reference semantic features.
By way of example, the reference session data set may be constructed in the following manner: acquiring an initial session data set comprising a plurality of reference session data; performing cluster analysis on each reference session data in the initial session data set; and respectively selecting different types of reference session data from the initial session data set to construct a reference session data set.
By way of example, the candidate policy dataset may be constructed in the following manner: acquiring an initial policy data set comprising a plurality of candidate policy data; performing cluster analysis on each candidate strategy data in the initial strategy data set; and respectively selecting candidate strategy data of different categories from the initial strategy data set, and constructing a candidate strategy data set.
Illustratively, the policy recall model is trained in the following manner: acquiring a sample session set; the sample session set comprises a plurality of sample session data and labeling strategy data corresponding to each sample session data; extracting features of the sample session data to obtain sample session features; inputting the sample session characteristics into a pre-constructed strategy recall model to obtain recall strategy data of corresponding sample session data; counting the duty ratio of the labeling strategy data with the maximum recall probability corresponding to the labeling strategy data of each sample session data in the candidate strategy data set, and taking the duty ratio as a first recall accuracy; determining a second recall accuracy according to the ranking order of recall probabilities of the labeling strategy data of each sample session data in the candidate strategy data set; a recall accuracy rate is generated that includes the first recall accuracy rate and/or the second recall accuracy rate. Determining the strategy accuracy of the sample session set according to the consistency of recall strategy data of each sample session data and corresponding labeling strategy data; and performing model training on the strategy recall model according to the recall accuracy and/or the strategy accuracy.
It should be noted that the above-mentioned computing device for performing different model training, the computing device for performing correspondence construction, and the device for performing the operation and maintenance policy determining method may be the same or at least partially different, which is not limited in this application.
It should be understood that, although the steps in the flowcharts related to the above embodiments are sequentially shown as indicated by arrows, these steps are not necessarily sequentially performed in the order indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in the flowcharts described in the above embodiments may include a plurality of steps or a plurality of stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily performed sequentially, but may be performed alternately or alternately with at least some of the other steps or stages.
Based on the same inventive concept, the embodiment of the application also provides an operation and maintenance policy determining device for implementing the above related operation and maintenance policy determining method. The implementation of the solution provided by the device is similar to the implementation described in the above method, so the specific limitation in the embodiments of the one or more operation and maintenance policy determining devices provided below may refer to the limitation of the operation and maintenance policy determining method hereinabove, and will not be described herein.
In an exemplary embodiment, an operation and maintenance policy determining apparatus as shown in fig. 6 includes: an operation and maintenance session data acquisition module 610, an operation and maintenance semantic feature extraction module 620 and a target operation and maintenance policy selection module 630. Wherein,
an operation and maintenance session data obtaining module 610, configured to obtain at least one operation and maintenance session data;
the operation and maintenance semantic feature extraction module 620 is configured to perform feature extraction on the operation and maintenance session data to obtain operation and maintenance semantic features;
the target operation and maintenance policy selecting module 630 is configured to select a target operation and maintenance policy from at least one candidate operation and maintenance policy according to the operation and maintenance semantic feature; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system.
According to the method and the device for processing the operation and maintenance session data, the operation and maintenance semantic features are obtained through feature extraction of the at least one operation and maintenance session data, and the target operation and maintenance policy is automatically selected from at least one candidate operation and maintenance policy for representing an operation and maintenance processing mode of the system according to the obtained operation and maintenance semantic features, so that the determination efficiency of the operation and maintenance policy is improved. Meanwhile, professional operation and maintenance personnel are not required to be introduced in the process to determine the operation and maintenance strategy, so that the labor cost of the operation and maintenance process is reduced, the professional threshold of the operation and maintenance personnel is reduced, and the influence of human factors on the accuracy of the operation and maintenance strategy determination result is avoided. In addition, the operation and maintenance semantic features obtained by feature extraction are based on at least one operation and maintenance session data, and the target operation and maintenance strategy is determined, so that the influence caused by different operation and maintenance session data can be comprehensively considered in the process of determining the target operation and maintenance strategy, the richness of data referenced in the process of determining the operation and maintenance strategy is improved, and the accuracy of the determination result of the operation and maintenance strategy is further improved.
In one embodiment, the operation and maintenance semantic feature extraction module 620 includes: the semantic association feature extraction unit is used for determining semantic association conditions among the session data without operation and maintenance and extracting semantic association features among the session data with different operation and maintenance according to the semantic association conditions; and the operation and maintenance semantic feature determining unit is used for taking the semantic association features as operation and maintenance semantic features.
In one embodiment, the semantic association feature extraction unit includes: the semantic association feature extraction subunit is used for determining semantic association conditions among the session data without operation and maintenance based on the trained language characterization model, and extracting semantic association features among the session data with different operation and maintenance according to the semantic association conditions;
the language characterization model is obtained by training a language characterization model training module, and the language characterization model training module comprises: the training session set acquisition unit is used for acquiring a training session set; wherein the training session set comprises a plurality of training session data; the training sample pair construction unit is used for constructing a plurality of training sample pairs comprising two training session data according to different training session data and semantic association labels among the different training session data; and the parameter fine tuning unit is used for carrying out parameter fine tuning on the pre-trained language characterization model according to the plurality of training sample pairs and the semantic association condition of each training sample pair.
In one embodiment, the training sample pairs include a training positive sample pair and a training negative sample pair; correspondingly, the training sample pair construction unit comprises: the positive sample pair constructing subunit is used for constructing a training positive sample pair according to any two training session data with semantic association; the negative sample pair constructing subunit is used for constructing a training negative sample pair according to any two training session data without semantic association.
In one embodiment, the parameter tuning unit includes: the covering processing subunit is used for covering the word segmentation result of the preset proportion in any training sample pair so as to update the training sample pair; the data input subunit is used for inputting the updated training sample pair into the pre-trained language characterization model to obtain sample semantic features; the category prediction subunit is used for predicting semantic association categories of different training session data in the training sample pair according to the sample semantic features; and the parameter fine tuning subunit is used for carrying out parameter fine tuning on the pre-trained language characterization model according to the semantic association category and the actual semantic association condition of the training sample pair.
In one embodiment, the target operation and maintenance policy selection module 630 includes: the corresponding relation acquisition unit is used for acquiring a preset first corresponding relation; the first corresponding relation is a corresponding relation between different reference semantic features and different candidate operation and maintenance strategies; the corresponding relation using unit is used for selecting a target operation and maintenance strategy corresponding to the operation and maintenance semantic feature from at least one candidate operation and maintenance strategy according to the first corresponding relation.
In one embodiment, the first correspondence is constructed by a correspondence construction module, where the opposite correspondence construction module includes: a data set acquisition unit configured to acquire a reference session data set including a plurality of reference session data, and a candidate policy data set including a plurality of candidate policy data; the feature extraction unit is used for respectively extracting features of each reference session data to obtain corresponding reference semantic features; the data recall unit is used for recalling candidate strategy data from the candidate strategy data set according to the reference semantic features; the corresponding relation establishing unit is used for establishing a first corresponding relation between the reference semantic feature and the recalled candidate strategy data corresponding to the reference semantic feature.
In one embodiment, a data recall unit, comprising: and the data recall subunit is used for recalling the candidate strategy data from the candidate strategy data set according to the reference semantic features based on the strategy recall model.
The strategy recall model is obtained by training a strategy recall model training module, and the strategy recall model training module comprises: a session set acquisition unit for acquiring a sample session set; the sample session set comprises a plurality of sample session data and labeling strategy data corresponding to each sample session data; the feature extraction unit is used for extracting features of the sample session data to obtain sample session features; the feature input unit is used for inputting the sample session features into a pre-constructed strategy recall model to obtain recall strategy data of corresponding sample session data; and the model training unit is used for carrying out model training on the strategy recall model according to the recall strategy data of each sample session data and the corresponding labeling strategy data.
In one embodiment, a model training unit includes: the recall accuracy determining subunit is used for determining the recall accuracy of the sample session set according to the ranking order of the recall probabilities corresponding to the labeling strategy data of each sample session data in the candidate strategy data set; the strategy accuracy rate determining subunit is used for determining the strategy accuracy rate of the sample session set according to the consistency condition of recall strategy data of each sample session data and corresponding labeling strategy data; and the model training subunit is used for carrying out model training on the strategy recall model according to the recall accuracy and/or the strategy accuracy.
In one embodiment, the recall accuracy determination subunit is specifically configured to: counting the maximum recall probability average value of the labeling strategy data with the maximum recall probability corresponding to the labeling strategy data of each sample session data in the candidate strategy data set, and taking the average value determination result as a first recall accuracy; determining a second recall accuracy according to the ranking order of recall probabilities of the labeling strategy data of each sample session data in the candidate strategy data set; a recall accuracy rate is generated that includes the first recall accuracy rate and/or the second recall accuracy rate.
In one embodiment, the data set obtaining unit in the correspondence building module includes a first data set obtaining subunit, specifically configured to: acquiring an initial session data set comprising a plurality of reference session data; performing cluster analysis on each reference session data in the initial session data set; and respectively selecting different types of reference session data from the initial session data set to construct a reference session data set.
In one embodiment, the data set obtaining unit in the correspondence building module includes a second data set obtaining subunit, specifically configured to: acquiring an initial policy data set comprising a plurality of candidate policy data; performing cluster analysis on each candidate strategy data in the initial strategy data set; and respectively selecting candidate strategy data of different categories from the initial strategy data set, and constructing a candidate strategy data set.
The above-mentioned respective modules in the operation and maintenance policy determining device may be implemented in whole or in part by software, hardware, and a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
In one exemplary embodiment, a robot, which may be a terminal, is provided, and an internal structure thereof may be as shown in fig. 7. The robot includes a processor, a memory, an input/output interface, a communication interface, a display unit, and an input device. The processor, the memory and the input/output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input/output interface. Wherein the processor of the robot is adapted to provide computing and control capabilities. The memory of the robot includes a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input/output interface of the robot is used for exchanging information between the processor and the external device. The communication interface of the robot is used for carrying out wired or wireless communication with an external terminal, and the wireless mode can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. The computer program is executed by a processor to implement an operation and maintenance policy determination method. The display unit of the robot is used for forming a visual picture and can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the robot can be a touch layer covered on the display screen, can also be keys, a track ball or a touch pad arranged on a robot shell, and can also be an external keyboard, a touch pad or a mouse and the like.
It will be appreciated by those skilled in the art that the structure shown in fig. 7 is merely a block diagram of some of the structures associated with the present application and is not limiting of the computer device to which the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.
In one exemplary embodiment, a computer device is provided comprising a memory and a processor, the memory having stored therein a computer program, the processor when executing the computer program performing the steps of: acquiring at least one operation and maintenance session data; extracting features of the operation and maintenance session data to obtain operation and maintenance semantic features; selecting a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance semantic features; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system.
In one embodiment, the processor when executing the computer program further performs the steps of: determining semantic association conditions among the operation and maintenance-free session data, and extracting semantic association features among different operation and maintenance session data according to the semantic association conditions; and taking the semantic association features as operation and maintenance semantic features.
In one embodiment, the processor when executing the computer program further performs the steps of: based on the trained language characterization model, determining semantic association conditions among the unnecessary operation and maintenance session data, and extracting semantic association features among different operation and maintenance session data according to the semantic association conditions.
In one embodiment, the processor when executing the computer program further performs the steps of: acquiring a training session set; wherein the training session set comprises a plurality of training session data; according to different training session data and semantic association labels among different training session data, constructing a plurality of training sample pairs comprising two training session data; and carrying out parameter fine adjustment on the pre-trained language characterization model according to the plurality of training sample pairs and the semantic association condition of each training sample pair.
In one embodiment, the training sample pairs include a training positive sample pair and a training negative sample pair; accordingly, the processor when executing the computer program also performs the steps of: according to any two training session data with semantic association, a training positive sample pair is constructed; and constructing a training negative sample pair according to any two training session data without semantic association.
In one embodiment, the processor when executing the computer program further performs the steps of: aiming at any training sample pair, masking word segmentation results with preset proportion in the training sample pair so as to update the training sample pair; inputting the updated training sample pair into a pre-trained language characterization model to obtain sample semantic features; predicting semantic association categories of different training session data in the training sample pair according to the sample semantic features; and carrying out parameter fine adjustment on the pre-trained language characterization model according to the semantic association category and the actual semantic association condition of the training sample pair.
In one embodiment, the processor when executing the computer program further performs the steps of: acquiring a preset first corresponding relation; the first corresponding relation is a corresponding relation between different reference semantic features and different candidate operation and maintenance strategies; and selecting a target operation and maintenance strategy corresponding to the operation and maintenance semantic features from at least one candidate operation and maintenance strategy according to the first corresponding relation.
In one embodiment, the processor when executing the computer program further performs the steps of: acquiring a reference session data set comprising a plurality of reference session data, and a candidate policy data set comprising a plurality of candidate policy data; respectively extracting features of each reference session data to obtain corresponding reference semantic features; recall candidate policy data from the candidate policy data set according to the reference semantic features; and establishing a first corresponding relation between the reference semantic features and the recalled candidate strategy data corresponding to the reference semantic features.
In one embodiment, the processor when executing the computer program further performs the steps of: based on the policy recall model, recall candidate policy data from the candidate policy data set according to the reference semantic features.
In one embodiment, the processor when executing the computer program further performs the steps of: acquiring a sample session set; the sample session set comprises a plurality of sample session data and labeling strategy data corresponding to each sample session data; extracting features of the sample session data to obtain sample session features; inputting the sample session characteristics into a pre-constructed strategy recall model to obtain recall strategy data of corresponding sample session data; and carrying out model training on the strategy recall model according to recall strategy data of each sample session data and corresponding labeling strategy data.
In one embodiment, the processor when executing the computer program further performs the steps of: determining recall accuracy of the sample session set according to the ranking order of recall probabilities corresponding to the labeling strategy data of each sample session data in the candidate strategy data set; determining the strategy accuracy of the sample session set according to the consistency of recall strategy data of each sample session data and corresponding labeling strategy data; and performing model training on the strategy recall model according to the recall accuracy and/or the strategy accuracy.
In one embodiment, the processor when executing the computer program further performs the steps of: counting the maximum recall probability average value of the labeling strategy data with the maximum recall probability corresponding to the labeling strategy data of each sample session data in the candidate strategy data set, and taking the average value determination result as a first recall accuracy; determining a second recall accuracy according to the ranking order of recall probabilities of the labeling strategy data of each sample session data in the candidate strategy data set; a recall accuracy rate is generated that includes the first recall accuracy rate and/or the second recall accuracy rate.
In one embodiment, the processor when executing the computer program further performs the steps of: acquiring an initial session data set comprising a plurality of reference session data; performing cluster analysis on each reference session data in the initial session data set; and respectively selecting different types of reference session data from the initial session data set to construct a reference session data set.
In one embodiment, the processor when executing the computer program further performs the steps of: acquiring an initial policy data set comprising a plurality of candidate policy data; performing cluster analysis on each candidate strategy data in the initial strategy data set; and respectively selecting candidate strategy data of different categories from the initial strategy data set, and constructing a candidate strategy data set.
In one embodiment, a computer readable storage medium is provided having a computer program stored thereon, which when executed by a processor, performs the steps of: acquiring at least one operation and maintenance session data; extracting features of the operation and maintenance session data to obtain operation and maintenance semantic features; selecting a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance semantic features; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system.
In one embodiment, the computer program when executed by a processor performs the steps of: determining semantic association conditions among the operation and maintenance-free session data, and extracting semantic association features among different operation and maintenance session data according to the semantic association conditions; and taking the semantic association features as operation and maintenance semantic features.
In one embodiment, the processor when executing the computer program further performs the steps of: based on the trained language characterization model, determining semantic association conditions among the unnecessary operation and maintenance session data, and extracting semantic association features among different operation and maintenance session data according to the semantic association conditions.
In one embodiment, the computer program when executed by a processor performs the steps of: acquiring a training session set; wherein the training session set comprises a plurality of training session data; according to different training session data and semantic association labels among different training session data, constructing a plurality of training sample pairs comprising two training session data; and carrying out parameter fine adjustment on the pre-trained language characterization model according to the plurality of training sample pairs and the semantic association condition of each training sample pair.
In one embodiment, the training sample pairs include a training positive sample pair and a training negative sample pair; accordingly, the computer program when executed by the processor performs the steps of: according to any two training session data with semantic association, a training positive sample pair is constructed; and constructing a training negative sample pair according to any two training session data without semantic association.
In one embodiment, the computer program when executed by a processor performs the steps of: aiming at any training sample pair, masking word segmentation results with preset proportion in the training sample pair so as to update the training sample pair; inputting the updated training sample pair into a pre-trained language characterization model to obtain sample semantic features; predicting semantic association categories of different training session data in the training sample pair according to the sample semantic features; and carrying out parameter fine adjustment on the pre-trained language characterization model according to the semantic association category and the actual semantic association condition of the training sample pair.
In one embodiment, the computer program when executed by a processor performs the steps of: acquiring a preset first corresponding relation; the first corresponding relation is a corresponding relation between different reference semantic features and different candidate operation and maintenance strategies; and selecting a target operation and maintenance strategy corresponding to the operation and maintenance semantic features from at least one candidate operation and maintenance strategy according to the first corresponding relation.
In one embodiment, the computer program when executed by a processor performs the steps of: acquiring a reference session data set comprising a plurality of reference session data, and a candidate policy data set comprising a plurality of candidate policy data; respectively extracting features of each reference session data to obtain corresponding reference semantic features; recall candidate policy data from the candidate policy data set according to the reference semantic features; and establishing a first corresponding relation between the reference semantic features and the recalled candidate strategy data corresponding to the reference semantic features.
In one embodiment, the computer program when executed by a processor performs the steps of: based on the policy recall model, recall candidate policy data from the candidate policy data set according to the reference semantic features.
In one embodiment, the computer program when executed by a processor performs the steps of: acquiring a sample session set; the sample session set comprises a plurality of sample session data and labeling strategy data corresponding to each sample session data; extracting features of the sample session data to obtain sample session features; inputting the sample session characteristics into a pre-constructed strategy recall model to obtain recall strategy data of corresponding sample session data; and carrying out model training on the strategy recall model according to recall strategy data of each sample session data and corresponding labeling strategy data.
In one embodiment, the computer program when executed by a processor performs the steps of: determining recall accuracy of the sample session set according to the ranking order of recall probabilities corresponding to the labeling strategy data of each sample session data in the candidate strategy data set; determining the strategy accuracy of the sample session set according to the consistency of recall strategy data of each sample session data and corresponding labeling strategy data; and performing model training on the strategy recall model according to the recall accuracy and/or the strategy accuracy.
In one embodiment, the computer program when executed by a processor performs the steps of: counting the maximum recall probability average value of the labeling strategy data with the maximum recall probability corresponding to the labeling strategy data of each sample session data in the candidate strategy data set, and taking the average value determination result as a first recall accuracy; determining a second recall accuracy according to the ranking order of recall probabilities of the labeling strategy data of each sample session data in the candidate strategy data set; a recall accuracy rate is generated that includes the first recall accuracy rate and/or the second recall accuracy rate.
In one embodiment, the computer program when executed by a processor performs the steps of: acquiring an initial session data set comprising a plurality of reference session data; performing cluster analysis on each reference session data in the initial session data set; and respectively selecting different types of reference session data from the initial session data set to construct a reference session data set.
In one embodiment, the computer program when executed by a processor performs the steps of: acquiring an initial policy data set comprising a plurality of candidate policy data; performing cluster analysis on each candidate strategy data in the initial strategy data set; and respectively selecting candidate strategy data of different categories from the initial strategy data set, and constructing a candidate strategy data set.
In one embodiment, a computer program product is provided comprising a computer program which, when executed by a processor, performs the steps of: acquiring at least one operation and maintenance session data; extracting features of the operation and maintenance session data to obtain operation and maintenance semantic features; selecting a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance semantic features; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system.
In one embodiment, the computer program when executed by a processor performs the steps of: determining semantic association conditions among the operation and maintenance-free session data, and extracting semantic association features among different operation and maintenance session data according to the semantic association conditions; and taking the semantic association features as operation and maintenance semantic features.
In one embodiment, the processor when executing the computer program further performs the steps of: based on the trained language characterization model, determining semantic association conditions among the unnecessary operation and maintenance session data, and extracting semantic association features among different operation and maintenance session data according to the semantic association conditions.
In one embodiment, the computer program when executed by a processor performs the steps of: acquiring a training session set; wherein the training session set comprises a plurality of training session data; according to different training session data and semantic association labels among different training session data, constructing a plurality of training sample pairs comprising two training session data; and carrying out parameter fine adjustment on the pre-trained language characterization model according to the plurality of training sample pairs and the semantic association condition of each training sample pair.
In one embodiment, the training sample pairs include a training positive sample pair and a training negative sample pair; accordingly, the computer program when executed by the processor performs the steps of: according to any two training session data with semantic association, a training positive sample pair is constructed; and constructing a training negative sample pair according to any two training session data without semantic association.
In one embodiment, the computer program when executed by a processor performs the steps of: aiming at any training sample pair, masking word segmentation results with preset proportion in the training sample pair so as to update the training sample pair; inputting the updated training sample pair into a pre-trained language characterization model to obtain sample semantic features; predicting semantic association categories of different training session data in the training sample pair according to the sample semantic features; and carrying out parameter fine adjustment on the pre-trained language characterization model according to the semantic association category and the actual semantic association condition of the training sample pair.
In one embodiment, the computer program when executed by a processor performs the steps of: acquiring a preset first corresponding relation; the first corresponding relation is a corresponding relation between different reference semantic features and different candidate operation and maintenance strategies; and selecting a target operation and maintenance strategy corresponding to the operation and maintenance semantic features from at least one candidate operation and maintenance strategy according to the first corresponding relation.
In one embodiment, the computer program when executed by a processor performs the steps of: acquiring a reference session data set comprising a plurality of reference session data, and a candidate policy data set comprising a plurality of candidate policy data; respectively extracting features of each reference session data to obtain corresponding reference semantic features; recall candidate policy data from the candidate policy data set according to the reference semantic features; and establishing a first corresponding relation between the reference semantic features and the recalled candidate strategy data corresponding to the reference semantic features.
In one embodiment, the computer program when executed by a processor performs the steps of: based on the policy recall model, recall candidate policy data from the candidate policy data set according to the reference semantic features.
In one embodiment, the computer program when executed by a processor performs the steps of: acquiring a sample session set; the sample session set comprises a plurality of sample session data and labeling strategy data corresponding to each sample session data; extracting features of the sample session data to obtain sample session features; inputting the sample session characteristics into a pre-constructed strategy recall model to obtain recall strategy data of corresponding sample session data; and carrying out model training on the strategy recall model according to recall strategy data of each sample session data and corresponding labeling strategy data.
In one embodiment, the computer program when executed by a processor performs the steps of: determining recall accuracy of the sample session set according to the ranking order of recall probabilities corresponding to the labeling strategy data of each sample session data in the candidate strategy data set; determining the strategy accuracy of the sample session set according to the consistency of recall strategy data of each sample session data and corresponding labeling strategy data; and performing model training on the strategy recall model according to the recall accuracy and/or the strategy accuracy.
In one embodiment, the computer program when executed by a processor performs the steps of: counting the maximum recall probability average value of the labeling strategy data with the maximum recall probability corresponding to the labeling strategy data of each sample session data in the candidate strategy data set, and taking the average value determination result as a first recall accuracy; determining a second recall accuracy according to the ranking order of recall probabilities of the labeling strategy data of each sample session data in the candidate strategy data set; a recall accuracy rate is generated that includes the first recall accuracy rate and/or the second recall accuracy rate.
In one embodiment, the computer program when executed by a processor performs the steps of: acquiring an initial session data set comprising a plurality of reference session data; performing cluster analysis on each reference session data in the initial session data set; and respectively selecting different types of reference session data from the initial session data set to construct a reference session data set.
In one embodiment, the computer program when executed by a processor performs the steps of: acquiring an initial policy data set comprising a plurality of candidate policy data; performing cluster analysis on each candidate strategy data in the initial strategy data set; and respectively selecting candidate strategy data of different categories from the initial strategy data set, and constructing a candidate strategy data set.
It should be noted that, the data (including, but not limited to, data for analysis, stored data, presented data, etc.) referred to in the present application are all information and data authorized by the user or sufficiently authorized by each party, and the collection, use, and processing of the relevant data are required to meet the relevant regulations.
Those skilled in the art will appreciate that implementing all or part of the above-described methods in accordance with the embodiments may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. Any reference to memory, database, or other medium used in the various embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high density embedded nonvolatile Memory, resistive random access Memory (ReRAM), magnetic random access Memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric Memory (Ferroelectric Random Access Memory, FRAM), phase change Memory (Phase Change Memory, PCM), graphene Memory, and the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, and the like. By way of illustration, and not limitation, RAM can be in the form of a variety of forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), and the like. The databases referred to in the various embodiments provided herein may include at least one of relational databases and non-relational databases. The non-relational database may include, but is not limited to, a blockchain-based distributed database, and the like. The processors referred to in the embodiments provided herein may be general purpose processors, central processing units, graphics processors, digital signal processors, programmable logic units, quantum computing-based data processing logic units, etc., without being limited thereto.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The foregoing examples represent only a few embodiments of the present application, which are described in more detail and are not thereby to be construed as limiting the scope of the present application. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application shall be subject to the appended claims.

Claims (10)

1. An operation and maintenance strategy determining method is characterized by comprising the following steps:
acquiring at least one operation and maintenance session data;
extracting features of the operation and maintenance session data to obtain operation and maintenance semantic features;
selecting a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance semantic features; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system.
2. The method according to claim 1, wherein the feature extraction of the operation and maintenance session data to obtain operation and maintenance semantic features includes:
determining semantic association conditions among the operation and maintenance-free session data, and extracting semantic association features among different operation and maintenance session data according to the semantic association conditions;
and taking the semantic association features as the operation and maintenance semantic features.
3. The method according to claim 2, wherein the determining the semantic association condition between the session data without operation and maintenance and extracting the semantic association feature between the session data with different operation and maintenance according to the semantic association condition comprises:
based on a trained language characterization model, determining semantic association conditions among the unnecessary operation and maintenance session data, and extracting semantic association features among different operation and maintenance session data according to the semantic association conditions;
the language characterization model is trained by the following modes:
acquiring a training session set; wherein the training session set comprises a plurality of training session data;
according to different training session data and semantic association labels among different training session data, constructing a plurality of training sample pairs comprising two training session data;
And carrying out parameter fine adjustment on the pre-trained language characterization model according to the plurality of training sample pairs and the semantic association condition of each training sample pair.
4. A method according to any one of claims 1-3, wherein said selecting a target operation and maintenance policy from at least one candidate operation and maintenance policy based on said operation and maintenance semantic features comprises:
acquiring a preset first corresponding relation; the first corresponding relation is a corresponding relation between different reference semantic features and different candidate operation and maintenance strategies;
and selecting a target operation and maintenance strategy corresponding to the operation and maintenance semantic feature from at least one candidate operation and maintenance strategy according to the first corresponding relation.
5. The method of claim 4, wherein the first correspondence is constructed by:
acquiring a reference session data set comprising a plurality of reference session data, and a candidate policy data set comprising a plurality of candidate policy data;
respectively extracting features of each reference session data to obtain corresponding reference semantic features;
recall candidate policy data from the candidate policy data set according to the reference semantic features;
And establishing a first corresponding relation between the reference semantic features and recalled candidate strategy data corresponding to the reference semantic features.
6. The method of claim 5, wherein recalling candidate policy data from the candidate policy data set in accordance with the reference semantic features comprises:
recall candidate policy data from the candidate policy data set according to the reference semantic features based on a policy recall model;
the strategy recall model is trained by the following modes:
acquiring a sample session set; the sample session set comprises a plurality of sample session data and labeling strategy data corresponding to each sample session data;
extracting features of the sample session data to obtain sample session features;
inputting the sample session characteristics into a pre-constructed strategy recall model to obtain recall strategy data of corresponding sample session data;
and carrying out model training on the strategy recall model according to recall strategy data of each sample session data and corresponding labeling strategy data.
7. An operation and maintenance policy determining device, comprising:
the operation and maintenance session data acquisition module is used for acquiring at least one operation and maintenance session data;
The operation and maintenance semantic feature extraction module is used for extracting features of the operation and maintenance session data to obtain operation and maintenance semantic features;
the target operation and maintenance strategy selection module is used for selecting a target operation and maintenance strategy from at least one candidate operation and maintenance strategy according to the operation and maintenance semantic characteristics; the candidate operation and maintenance strategy is used for representing an operation and maintenance processing mode of the system.
8. A robot comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, carries out the steps of the method of any one of claims 1 to 6.
9. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.
10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.
CN202311386702.3A 2023-10-24 2023-10-24 Operation and maintenance strategy determining method, device, robot, storage medium and program product Pending CN117454146A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311386702.3A CN117454146A (en) 2023-10-24 2023-10-24 Operation and maintenance strategy determining method, device, robot, storage medium and program product

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311386702.3A CN117454146A (en) 2023-10-24 2023-10-24 Operation and maintenance strategy determining method, device, robot, storage medium and program product

Publications (1)

Publication Number Publication Date
CN117454146A true CN117454146A (en) 2024-01-26

Family

ID=89580994

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311386702.3A Pending CN117454146A (en) 2023-10-24 2023-10-24 Operation and maintenance strategy determining method, device, robot, storage medium and program product

Country Status (1)

Country Link
CN (1) CN117454146A (en)

Similar Documents

Publication Publication Date Title
Qi et al. Finding all you need: web APIs recommendation in web of things through keywords search
US20220121906A1 (en) Task-aware neural network architecture search
CN110119477B (en) Information pushing method, device and storage medium
CN110020022B (en) Data processing method, device, equipment and readable storage medium
US20220172260A1 (en) Method, apparatus, storage medium, and device for generating user profile
TW202009741A (en) Feature relationship recommendation method and apparatus, computing device and storage medium
CN112508609A (en) Crowd expansion prediction method, device, equipment and storage medium
CN111429204A (en) Hotel recommendation method, system, electronic equipment and storage medium
CN112131261B (en) Community query method and device based on community network and computer equipment
CN113239159A (en) Cross-modal retrieval method of videos and texts based on relational inference network
CN115994226A (en) Clustering model training system and method based on federal learning
CN111522926A (en) Text matching method, device, server and storage medium
CN110705279A (en) Vocabulary selection method and device and computer readable storage medium
CN112464106A (en) Object recommendation method and device
CN116956183A (en) Multimedia resource recommendation method, model training method, device and storage medium
CN111382232A (en) Question and answer information processing method and device and computer equipment
CN114169418B (en) Label recommendation model training method and device and label acquisition method and device
CN117454146A (en) Operation and maintenance strategy determining method, device, robot, storage medium and program product
CN113469752A (en) Content recommendation method and device, storage medium and electronic equipment
CN111444338A (en) Text processing device, storage medium and equipment
CN117708437B (en) Recommendation method and device for personalized content, electronic equipment and storage medium
CN114417944B (en) Recognition model training method and device, and user abnormal behavior recognition method and device
CN115455306B (en) Push model training method, information push device and storage medium
CN117407504A (en) Operation and maintenance processing method, apparatus, device, storage medium and program product
CN111309766B (en) Service feature construction method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination