WO2019210820A1 - 一种信息输出方法及装置 - Google Patents

一种信息输出方法及装置 Download PDF

Info

Publication number
WO2019210820A1
WO2019210820A1 PCT/CN2019/084814 CN2019084814W WO2019210820A1 WO 2019210820 A1 WO2019210820 A1 WO 2019210820A1 CN 2019084814 W CN2019084814 W CN 2019084814W WO 2019210820 A1 WO2019210820 A1 WO 2019210820A1
Authority
WO
WIPO (PCT)
Prior art keywords
word
semantic
vector
target data
description text
Prior art date
Application number
PCT/CN2019/084814
Other languages
English (en)
French (fr)
Inventor
饶思维
刘裕良
田光见
Original Assignee
华为技术有限公司
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 华为技术有限公司 filed Critical 华为技术有限公司
Publication of WO2019210820A1 publication Critical patent/WO2019210820A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/20Natural language analysis

Definitions

  • the present application relates to the field of communications technologies, and in particular, to an information output method and apparatus.
  • the first-line engineer collects data from the fault occurrence site to help analyze the cause of the fault. For example, collect key performance indicators (KPIs), device alarms, and device logs for a period of time before and after the network device failure occurs. Parameter data. And the first-line engineers will describe the fault phenomenon and get the fault description text. The first-line engineers will report the collected KPI and other data and fault description text to the operation and maintenance department in the form of a faulty work order.
  • KPIs key performance indicators
  • device alarms device alarms
  • device logs for a period of time before and after the network device failure occurs. Parameter data.
  • Parameter data the first-line engineers will describe the fault phenomenon and get the fault description text.
  • the first-line engineers will report the collected KPI and other data and fault description text to the operation and maintenance department in the form of a faulty work order.
  • the operation and maintenance engineer manually selects some KPI, equipment alarm, equipment log and other parameter data from the data collected by the first line. Further, abnormality detection and mutual authentication are performed on the selected data, thereby analyzing the root cause of the fault and providing guidance for repairing the faulty network device.
  • the fault detection method for selecting and analyzing the parameter data related to the fault description text from the parameter data such as the KPI, the device alarm, and the device log by manual manual is low in efficiency and cannot meet the increasing network demand.
  • a text having the same keyword as the fault description text is searched, and the fault view analysis is performed according to the related parameter data of the text.
  • the relevant texts and fault description texts that are highly relevant and can be used to assist in analyzing the cause of the fault may not have the same keywords. Therefore, the data associated with the fault description text associated with assisting in analyzing the cause of the fault cannot be accurately found by the existing method.
  • the present application provides an information output method and apparatus capable of automatically and accurately finding data related to a fault description text for assisting in analyzing a cause of a fault.
  • the present application provides an information output method, the method comprising: acquiring a fault description text, the fault description text is used to describe a fault occurring in a network; generating a semantic vector of the fault description text by using a semantic generation model; Relevant texts of the types of target data respectively correspond to semantic vectors, which are used to assist in analyzing the cause of the fault; calculating the correlation between the semantic vector of the fault description text and the semantic vector of the related text of each target data; Outputting first data, the first data is target data having the greatest correlation between the semantic vector of each target data and the semantic vector of the fault description text, or the first data is a semantic vector and a fault description text in each target data The relevance of the semantic vector is greater than the target data of the preset threshold.
  • the present application can accurately find the target data associated with the fault description text by comparing the semantic vector of the fault description text with the semantic vector of the related text of the target data.
  • the fault is described as “industry users are slow to access the Internet”.
  • the name of the key indicator for fault analysis that is analyzed in this application is “downstream bandwidth control packet loss ratio”. It can be seen that there is no element that can be matched and associated literally, and this application is precisely through the semantic analysis mining to learn the domain knowledge such as "the relationship between the speed of the Internet and the proportion of the packet loss". Analysis of the joint. Therefore, by implementing the method described in the first aspect, it is possible to automatically and accurately find out data related to the fault description text for assisting in analyzing the cause of the malfunction.
  • the semantic vector corresponding to the related texts of the plurality of types of target data may also be generated by the semantic generation model.
  • semantic vectors corresponding to the related texts of the plurality of target data are respectively saved; correspondingly, the specific implementation manners of the semantic vectors corresponding to the related texts of the plurality of target data are: obtaining related texts of the saved plurality of target data respectively Corresponding semantic vector.
  • the semantic vectors corresponding to the related texts of the plurality of target data may be generated and saved in advance, and after receiving the fault description text, the semantic vectors and faults corresponding to the saved related texts of the plurality of target data may be directly used.
  • the semantic vector describing the text is subjected to correlation calculation, so that the semantic vector corresponding to the related text of the plurality of target data is temporarily generated after the failure description text is received. It can be seen that by implementing the embodiment, it is advantageous to quickly calculate the correlation between the semantic vector of the fault description text and the semantic vector of the related text of each target data.
  • the semantic generation model is generated according to a training of a word vector matrix corresponding to a historical fault description text, where the word vector matrix includes a word vector corresponding to each word in the historical fault description text, and the word vector is used for Represents the semantics of a word.
  • the semantics of the text can be expressed more accurately.
  • the foregoing multiple types of target data include at least two types of key performance indicators, device alarms, and device logs.
  • the target data is a key performance indicator
  • the related text of the target data is a key.
  • the name of the performance indicator; when the target data is the device alarm, the related text of the target data is the identifier of the device alarm; when the target data is the device log, the related text of the target data is the content fragment of the device log.
  • the present application provides a training method for a semantic generation model, the method comprising: acquiring a set of word vectors corresponding to the training text, wherein the word vector included in the set of word vectors corresponds to the words in the training text one by one, The word vector is used to represent the semantics of the word; the historical fault description text is converted into a word vector matrix composed of at least one word vector according to the word vector set; the semantic generation model is obtained according to the word vector matrix training, and the semantic generation model is used to generate the text. Semantic vector.
  • the set of word vectors corresponding to the training text may be saved, so as to subsequently use the word vector in the set of word vectors.
  • the method described in the second aspect is to obtain a semantic generation model from the semantic level of the lexical level to the semantic level of the sentence level.
  • the semantic generation model training method conforms to the basic principle of language generation. Therefore, the semantic generation model trained by implementing the method described in the second aspect can more accurately express the semantics of the text.
  • the specific implementation manner of converting the historical fault description text into a word vector matrix composed of at least one word vector according to the word vector set is: performing word segmentation processing on the historical fault description text to obtain a historical fault description text.
  • Corresponding word sequence consisting of at least one word; obtaining a word vector corresponding to the word included in the word sequence from the word vector set; and forming a word vector matrix corresponding to the word vector corresponding to each word included in the word sequence.
  • the historical fault description text can be accurately converted into a word vector matrix composed of at least one word vector.
  • the random vector is generated as the word vector corresponding to the word included in the word sequence.
  • the historical fault description text can be accurately converted into a word vector matrix composed of at least one word vector.
  • the specific implementation manner of the semantic generation model according to the word vector matrix training is: acquiring a fault device type corresponding to the historical fault description text; and training the classification model according to the word vector matrix and the category label, the category label includes The faulty device type; the semantic generation model is obtained according to the classification model.
  • the semantics of the text can be expressed more accurately.
  • the specific implementation manner of training the classification model according to the word vector matrix and the category label is: inputting the word vector matrix and the category label into the neural network for iterative training, and inputting the neural network for each iteration training
  • the word vector in the word vector matrix and the parameters of the neural network are adjusted to generate the classification model.
  • the semantic generation model trained by this embodiment can more accurately express the semantics of the text.
  • the word vector in the word vector matrix input using the last iteration training may also update the word vector corresponding to the corresponding word in the word vector set.
  • the word vector in the text corpus correction word vector set can be described according to the historical fault with the domain knowledge, so that the word vector in the word vector set can more express the semantic information of the word of the domain knowledge.
  • an information output device that can perform the method of the first aspect or the possible embodiments of the first aspect described above.
  • This function can be implemented in hardware or in hardware by executing the corresponding software.
  • the hardware or software includes one or more units corresponding to the functions described above.
  • the unit can be software and/or hardware.
  • a model training device that can perform the method of the second aspect or the possible embodiments of the second aspect described above.
  • This function can be implemented in hardware or in hardware by executing the corresponding software.
  • the hardware or software includes one or more units corresponding to the functions described above.
  • the unit can be software and/or hardware.
  • an information output device comprising: a processor, a memory, a communication interface; a processor, a communication interface, and a memory connection; wherein the communication interface can be a transceiver.
  • the communication interface is used to implement communication with other network elements.
  • one or more programs are stored in a memory
  • the processor invoking a program stored in the memory to implement the solution in the first aspect or the possible implementation manner of the first aspect, the information output device solving the problem
  • a model training device comprising: a processor, a memory, a communication interface; a processor, a communication interface, and a memory connection; wherein the communication interface can be a transceiver.
  • the communication interface is used to implement communication with other network elements.
  • one or more programs are stored in a memory
  • the processor invoking a program stored in the memory to implement the solution in the second aspect or the possible implementation manner of the second aspect, the model training device solving the problem
  • a computer program product which, when run on a computer, causes the computer to perform the first aspect, the second aspect, the possible implementation of the first aspect, or the possible implementation of the second aspect Methods.
  • a chip product of an information output device the method of any of the first aspect or the first aspect of the first aspect.
  • a chip product of a model training device the method of any of the above-described second or second aspect of the second aspect.
  • a computer readable storage medium storing instructions for causing the computer to perform the method of the first aspect or the possible implementation of the first aspect when it is run on a computer The method in the way.
  • a computer readable storage medium storing instructions, when executed on a computer, causing the computer to perform the method of the second aspect or the second aspect The method in the embodiment.
  • FIG. 1 is a schematic flowchart of an information output method according to an embodiment of the present application
  • FIG. 2 is a schematic flowchart of a training method of a semantic generation model provided by an embodiment of the present application
  • FIG. 3 is a schematic diagram of a neural network used by a CBOW algorithm according to an embodiment of the present application
  • FIG. 4 is a schematic structural diagram of a neural network for training a classification model according to an embodiment of the present application
  • FIG. 5 is a schematic structural diagram of an information output apparatus according to an embodiment of the present disclosure.
  • FIG. 6 is a schematic structural diagram of a model training device according to an embodiment of the present application.
  • FIG. 7 is a schematic structural diagram of another information output apparatus according to an embodiment of the present application.
  • FIG. 8 is a schematic structural diagram of another model training apparatus according to an embodiment of the present application.
  • the embodiment of the present application provides an information output method and apparatus, which can automatically determine and output data related to the fault description text for assisting in analyzing the cause of the fault.
  • FIG. 1 is a schematic flowchart diagram of an information output method according to an embodiment of the present application. As shown in FIG. 1, the information output method includes the following 101 to 105 parts, wherein:
  • the information output device acquires a fault description text.
  • the fault description text is the text describing the fault phenomenon, that is, the fault description text is used to describe the fault that occurs in the network.
  • the fault description text may be “industry users are slow to access the Internet” and “online charging system (OCS) communication is interrupted”.
  • OCS online charging system
  • the fault description text may be sent by other devices to the information output device.
  • the first-line engineer describes the fault phenomenon, obtains the fault description text, and sends the collected data (such as key performance indicators, etc.) and the fault description text used to assist in analyzing the fault cause to the operation and maintenance department in the form of a fault work order.
  • Information output device is the text describing the fault phenomenon, that is, the fault description text is used to describe the fault that occurs in the network.
  • the fault description text may be “industry users are slow to access the Internet” and “online charging system (OCS) communication is interrupted”.
  • OCS online charging system
  • the fault description text may be sent by other devices to the information output device.
  • the first-line engineer describes
  • the information output device generates a semantic vector of the fault description text by using a semantic generation model.
  • the semantic generation model may be generated according to a training of a word vector matrix corresponding to the historical fault description text, where the word vector matrix includes a word vector corresponding to each word in the historical fault description text.
  • the training method of the semantic generation model may be specifically referred to the training method of the semantic generation model described in FIG. 2 below.
  • the semantic generation model used by the information output device may be a semantic generation model trained by the model training device in FIG. 2 below.
  • the information output device of FIG. 1 and the model training device of FIG. 2 may be deployed on the same device or deployed on different devices.
  • the model training device may send the semantic generation model to the information output device after training the semantic generation model, so that the information output device may receive
  • the semantic generation model generates a semantic vector of the fault description text.
  • the information output device in FIG. 1 and the model training device in FIG. 2 are deployed in the same device, the information output device can acquire a semantic generation model from the model training device, so that the information output device can generate a fault description through the semantic generation model.
  • the semantic vector of the text may be specifically referred to the training method of the semantic generation model described in FIG. 2 below. That is to say, the semantic generation model used by the information output device may be a semantic generation model
  • semantic generation model may also be generated by the training in the manner described in FIG. 2, and may be generated by other methods.
  • the specific implementation manner of the information output device generating the semantic vector of the fault description text by using the semantic generation model is:
  • the information output device converts the fault description text into a word vector matrix according to the word vector set, and then inputs the word vector matrix into the semantic generation model to generate a semantic vector of the fault description text, wherein the word vector set includes a plurality of word vectors.
  • the word vector set may be generated by the model training device in FIG. 2 and sent to the information output device.
  • the information output device converts the fault description text into a word vector matrix according to the word vector set: the information output device performs word segmentation processing on the fault description text, and obtains a word composed of at least one word corresponding to the fault description text. a sequence; a word vector corresponding to the word included in the word sequence is obtained from the word vector set; and the word vector corresponding to each word included in the word sequence constitutes a word vector matrix of the fault description text.
  • a random vector is generated as the word vector corresponding to the word included in the word sequence.
  • the fault description text includes 4 words, and the word sequence obtained by performing word segmentation on the fault description text is “industry”, “user”, “online”, and “slow”.
  • the information output device finds the "industry” corresponding word vector 1, the "user” corresponding word vector 2, the "online” corresponding word vector 3, and the "slow” corresponding word vector is not found from the word vector set, and generates a random vector word.
  • Vector 4 is used as the word vector corresponding to "slow”.
  • the information output device groups the word vectors 1 to 4 into a word vector matrix of the fault description text.
  • the word vector matrix is then input into a semantic vector of the fault description text in the semantic generation model.
  • the information output device acquires a semantic vector corresponding to each of the related texts of the plurality of types of target data.
  • the target data is used to assist in analyzing the cause of the fault.
  • the order of execution of the 103 part and the 102 part may be in no particular order, and the part 102 may be executed first and then the part 103 may be executed first, or the part 103 may be executed first.
  • the multiple types of target data include at least two of a key performance indicator (KPI), a device alarm, and a device log; and when the target data is a key performance indicator, the related text of the target data The name of the key performance indicator.
  • KPI key performance indicator
  • the target data is the device alarm
  • the related text of the target data is the identifier of the device alarm.
  • the target data is the device log
  • the related text of the target data is the content fragment of the device log.
  • each type of target data is multiple.
  • the multiple types of target data include key performance indicators and device alerts.
  • the above-mentioned multiple types of target data are 100 different key performance indicators and 20 different device alarms, and 100 key performance indicators are key performance indicators 1 to key performance indicators 100.
  • 20 device alarms are device alarms 1 respectively.
  • the semantic vectors corresponding to the related texts of the plurality of types of target data acquired by the information output device are the semantic vectors corresponding to the names of the key performance indicators 1 to the key performance indicators 100, and the semantic vectors corresponding to the identifiers of the device alarms 1 to 20 respectively. . That is to say, the information output device acquires 120 semantic vectors.
  • the information output device may generate a semantic vector corresponding to the related text of the plurality of types of target data by using the semantic generation model before receiving the fault description text.
  • the semantic vectors corresponding to the related texts of the plurality of types of target data may be saved.
  • a semantic vector corresponding to the related text storing the plurality of target data may be acquired to perform correlation calculation with the semantic vector of the fault description text.
  • the semantic vectors corresponding to the related texts of the plurality of target data may be generated and saved in advance, and after receiving the fault description text, the semantic vectors and faults corresponding to the saved related texts of the plurality of target data may be directly used.
  • the semantic vector describing the text is subjected to correlation calculation, so that the semantic vector corresponding to the related text of the plurality of target data is temporarily generated after the failure description text is received. It can be seen that by implementing the embodiment, it is advantageous to quickly calculate the correlation between the semantic vector of the fault description text and the semantic vector of the related text of each target data.
  • the principle that the information output device generates the semantic vector corresponding to the related text of the target data by using the semantic generation model is the same as the principle that the information output device generates the semantic vector of the fault description text by using the semantic generation model, and Narration.
  • the information output device calculates a correlation between a semantic vector of the fault description text and a semantic vector of the related text of each target data.
  • target data there are two types of target data, namely 100 different key performance indicators and 20 different device alarms, and 100 key performance indicators are key performance indicators 1 to key performance indicators 100.
  • 20 device alarms The alarms are 1 to 20 for the device.
  • the information output device calculates the correlation between the semantic vector of the fault description text and the semantic vector of the related text of the 100 key performance indicators, and the correlation between the semantic vector of the fault description text and the semantic vector of the related text of the 20 device alarms, respectively. . Therefore, there will be 120 correlations.
  • the angle of the vector may be used as a measure of correlation
  • the correlation between the semantic vector of the fault description text and the semantic vector of the related text of the target data may be expressed as:
  • cos( ⁇ ) is the correlation between the semantic vector of the fault description text and the semantic vector of the related text of the target data
  • n is the number of dimensions of the semantic vector of the fault description text and the related text of the target data
  • x i is the fault description text
  • the information output device determines and outputs the first data.
  • the information output device determines and outputs the first data, the first data is The target data in which the semantic vector of each target data has the greatest correlation with the semantic vector of the fault description text, or the first data is that the semantic vector of the semantic vector and the fault description text in each target data is greater than a preset threshold. Target data.
  • the two types of target data obtained are 100 different key performance indicators and 20 different device alarms.
  • the 100 key performance indicators are key performance indicators 1 to key performance indicators 100.
  • 20 device alarms are respectively Alarm 1 to device alarm 20.
  • the correlation between the semantic vector of the fault description text and the semantic vector of the related text of the key performance indicator 1 to the key performance indicator 100 is a correlation of 1 to 100, respectively.
  • Correlation 1 is the maximum correlation, and the information output device outputs the key performance indicator 1.
  • the correlation between the semantic vector of the fault description text and the semantic vector of the related text of the device alarm 1 to the device alarm 20 is the correlation 101 to 120, respectively.
  • the correlation 120 is the maximum correlation, and the information output device outputs the device alarm 20.
  • the information output device outputs the key performance indicator 1 and the key performance indicator 2.
  • the correlation 101 and the correlation 102 are correlations greater than a preset threshold, and the information output device outputs the device alarm 1 and the device alarm 2.
  • the target data with higher correlation between the semantic vector and the semantic vector of the fault description text indicates that the target data is more related to the fault description text, and the user may need to view the target data to analyze the cause of the fault.
  • the fault description text is “ocs communication interruption”
  • the key indicator name is “ocs communication interruption number”.
  • the semantic vector of the fault description text has a great correlation with the semantic vector of the name of the key indicator, and the user may need to view This key indicator is used to analyze the cause of the failure. It can be seen that by implementing the method described in FIG. 1, data related to the fault description text can be automatically found to assist in analyzing the cause of the fault.
  • the embodiment of the present application can accurately find the target data associated with the fault description text by comparing the semantic vector of the fault description text with the semantic vector of the related text of the target data.
  • the fault is described as “the industry user is slow to access the Internet.”
  • the key indicator for fault analysis that is analyzed in the embodiment of the present application is “downstream bandwidth control packet loss ratio”. It can be seen that there is no element that can be matched and associated literally, and this application is precisely through the semantic analysis mining to learn the domain knowledge such as "the relationship between the speed of the Internet and the proportion of the packet loss”. Analysis of the joint.
  • FIG. 2 is a schematic flowchart diagram of a training method of a semantic generation model according to an embodiment of the present application.
  • the training method of the semantic generation model includes the following sections 201-203, wherein:
  • the model training device acquires a set of word vectors corresponding to the training text.
  • the word vector included in the word vector set corresponds one-to-one with the words in the training text. For example, if the training text includes 10,000 words, then the word vector set also includes 10000 word vectors. This word vector is used to represent the semantics of the word.
  • the set of word vectors corresponding to the training text may be saved, so as to subsequently use the word vector in the set of word vectors.
  • the training text is the corpus.
  • the training text may be an encyclopedic text. Word vectors learned from encyclopedic texts have good general semantics.
  • the model training device first preprocesses the training text, divides the sentence according to the sentence, and then performs word segmentation processing on each sentence text, obtains the training text after the word segmentation, and obtains the word segmentation through the word2vec tool or other tools.
  • a collection of word vectors corresponding to the training text is first preprocessed.
  • the training text is "Mathematics is a discipline that uses symbolic language to study quantitative structure changes and concepts such as space. I like mathematics.”
  • the model training device splits the training text into two sentences, namely, "Mathematics is a discipline that uses symbolic language to study quantitative structure changes and space and other concepts" and "I like mathematics". Then separate the two sentences into word segmentation.
  • the training text after the word segmentation is "Mathematics is a discipline that uses symbolic language to study quantitative structure changes and concepts such as space. I like mathematics.”
  • the model training device uses the word2vec tool to traverse the training text after the word segmentation step by step, and the traversal ends to obtain the word vector corresponding to each word in the training text.
  • the model training device saves a set of word vectors composed of word vectors corresponding to each word in the training text.
  • the model training device can obtain the set of word vectors corresponding to the training text after the word segmentation by using the word2vec tool and using the CBOW algorithm.
  • the idea of the CBOW algorithm is to predict the current word by a given context word.
  • the goal of CBOW algorithm training is to give a maximum probability of occurrence of a word when it is given. After the training, each word gets a corresponding word vector in the output layer.
  • the modeling idea of the CBOW algorithm is a classification process, it generates a by-product of the word vector.
  • Figure 3 is a schematic diagram of a neural network employed by the CBOW algorithm.
  • the neural network is composed of a three-layer structure, which is an input layer, a mapping layer, and an output layer.
  • the output layer includes a Huffman tree that has been constructed.
  • a leaf node of the Huffman tree represents a word vector of a word in the training text, and the word vector of the word corresponding to each leaf node is randomly initialized.
  • Each non-leaf node has a weight vector built in, the dimension of which is the same as the word vector of the input layer.
  • the input layer is a word vector of n-1 words around a certain word w(t).
  • n is the window size.
  • the n-1 words around the word w(t) are the first two and the last two words of the word w(t).
  • the first two and the last two words of the word w(t) are w(t-2), w(t-1), w(t+1), and w(t+2), respectively.
  • the word vectors of the n-1 words are denoted as v(w(t-2)), v(w(t-1)), v(w(t+1)), v(w(t +2)).
  • the input layer passes the n-1 word vectors to the mapping layer, and the mapping layer adds n-1 word vectors, that is, the respective dimensions of the n-1 word vectors are added.
  • the projection layer inputs the summed vector pro(t) into the root node of the Huffman tree. After the pro(t) is input to the root node, the probability of the root node to each leaf node is calculated.
  • the training process of the model is expected to obtain the maximum probability of reaching the leaf node corresponding to w(t) by the root node, due to the massive amount In the training text, the same context will appear multiple times, so the weight vector will be continually corrected during the process of traversing the training text to achieve such an effect.
  • the word vector corresponding to each leaf node of the Huffman tree is the word vector corresponding to each word of the training text.
  • "all words in the training text" includes repeated words in the training text.
  • the leaf node corresponding to the word w(t) from the root node to the word w(t) is equivalent to one secondary classification each time, and the classifier can adopt the softmax regression classifier.
  • the classification probability of each classification is:
  • ⁇ i represents the ith weight vector.
  • Pro(t) is the sum of the word vectors of the context of w(t), and e is a natural constant.
  • the model training device converts the historical fault description text into a word vector matrix composed of at least one word vector according to the set of word vectors.
  • the model training device can convert a large amount of historical fault description text into a word vector matrix.
  • the model training device trains a semantic generation model based on a large number of word vector matrix training. For example, with the historical fault description text 1 - the historical fault description text 100, the historical fault description text 1 - the historical fault description text 100 can be converted into a word vector matrix, that is, 100 word vector matrices are obtained.
  • the model training device obtains a semantic generation model based on the training of the 100 word vector matrices.
  • the model training device converts the historical fault description text into a word vector matrix composed of at least one word vector according to the set of word vectors, and the model training device may perform word segmentation on the historical fault description text. Processing, obtaining a sequence of words consisting of at least one word corresponding to the historical fault description text; obtaining a word vector corresponding to the word included in the sequence of words from the set of word vectors; forming a word vector corresponding to each word included in the sequence of words The word vector matrix of the fault description text. When there is no word vector corresponding to the word included in the word sequence in the word vector set, a random vector may be generated as the word vector corresponding to the word included in the word sequence. It can be seen that by implementing this embodiment, the historical fault description text can be accurately converted into a word vector matrix composed of at least one word vector.
  • the historical fault description text 1 includes 4 words, and the word sequence obtained by performing word segmentation on the historical fault description text 1 is “industry”, “user”, “online”, and “slow”.
  • the model training device finds the "industry” corresponding word vector 1, the "user” corresponding word vector 2, the "online” corresponding word vector 3, and the "slow” corresponding word vector is not found in the word vector set, and generates a random vector word.
  • Vector 4 is used as the word vector corresponding to "slow”.
  • the model training device composes the word vectors 1 to 4 into the word vector matrix 1 of the historical failure description text 1.
  • the principle of converting other historical fault description texts 2 to 100 into word vector matrix is the same as the principle of historical fault description text 1 converted into word vector matrix, and will not be described here.
  • the model training device obtains a semantic generation model according to the word vector matrix training.
  • the model training device may input the word vector matrix into the neural network for training to obtain a semantic generation model.
  • This semantic generation model is used to generate semantic vectors for text.
  • This semantic vector is used to represent the semantics of the text.
  • the method described in FIG. 2 is a semantic generation model from the semantic level of the lexical level to the semantic level of the sentence level.
  • the semantic generation model training method conforms to the basic principle of language generation. Therefore, the semantic generation model trained by implementing the method described in FIG. 2 can more accurately express the semantics of the text.
  • the model training device obtains the semantic generation model according to the word vector matrix training: the model training device acquires the fault device type corresponding to the historical fault description text; and the model training device according to the word vector matrix and the category
  • the tag training classification model includes the faulty device type; the model training device obtains a semantic generation model according to the classification model.
  • the faulty device type corresponding to the historical fault description text may be a router, a wired device, or a wireless device.
  • the faulty device type corresponding to the historical fault description text is a router.
  • the first-line engineer can collect the faulty device type corresponding to each fault description text, and then add the fault description text, the faulty device type corresponding to the fault description text, and the data used to assist in analyzing the cause of the fault to the work order, and send the work order to the work order.
  • the operation and maintenance terminal performs fault analysis. Therefore, the model training device can obtain the faulty device type corresponding to the historical fault description text from the work order.
  • the classification model obtained by the training is a model for generating a faulty device type corresponding to the fault description text.
  • the word vector matrix corresponding to the fault description text 1 is input into a classification model, and the classification model can output the fault device type corresponding to the fault description text 1.
  • the model training apparatus trains the classification model according to the word vector matrix and the category label, and the specific implementation manner is: inputting the word vector matrix and the category label into the neural network for iterative training, and inputting the input in each iteration training
  • the word vectors in the word vector matrix of the neural network and the parameters of the neural network are adjusted to generate a classification model.
  • the training classification model can be accurately classified into the fault description text.
  • the model training device may further update the word vector corresponding to the corresponding word in the word vector set by using the word vector in the adjusted word vector matrix.
  • the word vector in the text corpus correction word vector set can be described according to the historical fault with the domain knowledge, so that the word vector in the word vector set can more express the semantic information of the word in the fault domain.
  • FIG. 4 is a schematic structural diagram of a neural network for training a classification model.
  • the neural network includes a convolution layer, a pooling layer, and a fully connected layer.
  • the word vector matrix 1 of the historical fault description text 1 includes the word vectors ⁇ w1, w2, w3, w4, w5, w6 ⁇ .
  • the dimension of each word vector is 128 dimensions.
  • the model training device obtains the word vector matrix 1, the word vector matrix 1 is input to the neural network.
  • the convolution kernel 1 on the left performs a two-two convolution on the word vector matrix 1 including the word vector. For example, w1 and w2 are convoluted to obtain C1, w2 and w3 are convoluted to obtain C2, w3 and w4 are convoluted to obtain C3, w4 and w5 are convoluted to obtain C4, and w5 and w6 are convoluted to obtain C5.
  • the convolution kernel 2 on the right performs a three-three convolution on the word vector matrix 1 including the word vector.
  • w1, w2, and w3 are convoluted to obtain C6, w2, w3, and w4 are convoluted to obtain C7, w3, w4, and w5 are convoluted to obtain C8, and w4, w5, and w6 are convoluted to obtain C9.
  • Other numbers of word vectors are also convoluted in practical applications.
  • the embodiments of the present application are exemplified by two-two convolutions and three-three convolutions.
  • the model training device obtains the feature map generated by each convolution kernel, for each feature map, the maximum value in each dimension is selected as the text feature vector generated by the current convolution kernel by the maximum pooling operation.
  • the model training device splicing all the text feature vectors to obtain the semantic vector of the final historical fault description text 1. That is, as shown in FIG. 4, the model training device selects the largest value from the first dimension of C1 to C5, and selects the largest value from the second dimension of C1 to C5, from the third dimension of C1 to C5.
  • the model training device composes the maximum value of the selected 128 dimensions into the text feature vector 1 corresponding to the convolution kernel 1. Similarly, the model training device also acquires the text feature vector 2 corresponding to the convolution kernel 2. The model training device splices the text feature vector 1 and the text feature vector 2 to obtain a semantic vector of the final historical fault description text 1.
  • the model training device inputs the semantic vector of the obtained historical fault description text 1 into the fully connected layer, and inputs the faulty device type (such as a router) corresponding to the historical fault description text 1 as a category label, and inputs the fully connected layer.
  • the model training device analyzes the semantic vector of the historical fault description text 1 at the full connection layer, and analyzes the maximum probability of the faulty device type as the switch.
  • the faulty device type ie, the switch
  • the faulty device type whose maximum probability is obtained by analyzing the semantic vector of the historical fault description text 1 is different from the category label (ie, the router) corresponding to the historical fault description text 1, so the model training device records the history fault description.
  • the semantic vector of text 1 is analyzed to obtain the maximum probability of the faulty device type is incorrect.
  • the model training device inputs the word vector matrix of the historical fault description text 2 into the neural network according to the above process, and obtains the semantic vector of the historical fault description text 2, and inputs the fault device corresponding to the historical fault description text 2 at the full connection layer.
  • Types (such as switches) are used as category labels.
  • the model training device analyzes the semantic vector of the historical fault description text 2, and analyzes that the maximum probability of the faulty device type is the firewall. Therefore, the model training device records the faulty device type of the maximum probability obtained by analyzing the semantic vector of the historical fault description text 2 to be incorrect. Assume that there are 100 historical fault description texts, and the remaining 98 historical fault description texts are the same.
  • the neural network for training of the classification model according to the above-mentioned historical fault description text 1.
  • the fault device type of the maximum probability obtained by the semantic vector analysis corresponding to the historical fault description texts 1 to 50 is incorrect, and the parameters of the model training device for the neural network and The word vector in the word vector matrix corresponding to the historical fault description texts 1 to 50 is adjusted.
  • the historical fault description texts 1 to 100 are re-trained with the parameters of the new word vector matrix and the neural network until the faulty device type and classification of the maximum probability obtained by the semantic vector corresponding to the historical fault description texts 1 to 100 are analyzed.
  • a classification model is generated, that is, the classification model is generated by iteratively training the neural network.
  • the final model training device uses the word vector in the word vector matrix of the last round of iterative training to update the word vector corresponding to the corresponding word in the word vector set.
  • the historical fault description text 1 is "the slow Internet speed" before the last round of iterative training, the historical fault description text 1 corresponding word vector matrix is adjusted, and the word vector corresponding to "online” is adjusted to the word vector 1, the last time After the iterative training is completed, the word vector 1 is used to replace the word vector corresponding to "online” in the word vector set.
  • Historical fault description text 2 is the "OCS communication interruption" before the last round of iterative training, the historical fault description text 2 corresponding word vector matrix is adjusted, the word vector corresponding to "interruption” is adjusted to word vector 2, then the last iteration After the training is completed, the word vector 2 is used to replace the word vector corresponding to the "interrupt” in the word vector set.
  • Other historical fault description texts are similar, and will not be described here.
  • the specific implementation manner in which the model training device obtains the semantic generation model according to the classification model is: the model training device uses a portion above the fully connected layer in the classification model as a semantic generation model.
  • the semantic generation model generated by the embodiment the semantic vector of the text can be accurately generated.
  • the embodiment of the present invention may divide the function module into the device according to the foregoing method example.
  • each function module may be divided according to each function, or two or more functions may be integrated into one module.
  • the above integrated modules can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of the module in the embodiment of the present invention is schematic, and is only a logical function division, and the actual implementation may have another division manner.
  • FIG. 5 is an information output device provided by an implementation of the present invention.
  • the information output device includes an acquisition module 501, a generation module 502, a calculation module 503, and an output module 504. among them:
  • the obtaining module 501 is configured to obtain the fault description text, and the generating module 502 is configured to generate a semantic vector of the fault description text by using a semantic generation model, where the fault description text is used to describe a fault that occurs in the network, and the obtaining module 501 is further configured to obtain Correlation texts of the plurality of types of target data respectively correspond to semantic vectors for assisting in analyzing the cause of the fault; the calculation module 503 is configured to calculate the semantic vector of the fault description text and the semantics of the related text of each target data Correlation of the vector; the output module 504 is configured to determine and output the first data, where the first data is the target data with the greatest correlation between the semantic vector of each target data and the semantic vector of the fault description text, or the first data is The correlation between the semantic vector and the semantic vector of the fault description text in the target data is greater than the target data of the preset threshold.
  • the generating module 502 is further configured to generate a semantic vector corresponding to the related text of the plurality of target data by using the semantic generation model before the obtaining module 501 acquires the fault description text.
  • the semantic generation model is generated according to a training of a word vector matrix corresponding to the historical fault description text, and the word vector matrix includes a word vector corresponding to each word in the historical fault description text, and the word vector is used to represent the word.
  • the semantics are generated according to a training of a word vector matrix corresponding to the historical fault description text, and the word vector matrix includes a word vector corresponding to each word in the historical fault description text, and the word vector is used to represent the word.
  • the multiple types of target data include at least two of a key performance indicator, a device alarm, and a device log.
  • the target data is a key performance indicator
  • the related text of the target data is a key performance indicator.
  • the name of the target data is the identifier of the device alarm when the target data is the device alarm.
  • the target data is the device log
  • the related text of the target data is the content fragment of the device log.
  • FIG. 6 is a model training device provided by the implementation of the present invention.
  • the model training device includes an acquisition module 601, a conversion module 602, and a training module 603, wherein:
  • the obtaining module 601 is configured to obtain a set of word vectors corresponding to the training text, and the word vector included in the set of word vectors is in one-to-one correspondence with the words in the training text; and the converting module 602 is configured to convert the historical fault description text into a set according to the set of word vectors.
  • a word vector matrix composed of at least one word vector; the training module 603 is further configured to obtain a semantic generation model according to the word vector matrix training, and the semantic generation model is used to generate a semantic vector of the text.
  • the converting module 602 is specifically configured to: perform word segmentation processing on the historical fault description text, obtain a word sequence composed of at least one word corresponding to the historical fault description text, and obtain the word sequence from the word vector set.
  • the word vector corresponding to the word; the word vector corresponding to each word included in the word sequence constitutes a word vector matrix.
  • the conversion module 602 is further configured to: when there is no word vector corresponding to the word included in the word sequence in the word vector set, generate a random vector as the word vector corresponding to the word included in the word sequence.
  • the manner in which the training module 603 trains the semantic generation model according to the word vector matrix is specifically: acquiring the fault device type corresponding to the historical fault description text; and training the classification model according to the word vector matrix and the category label, the category
  • the tag includes the faulty device type; a semantic generation model is obtained according to the classification model.
  • the training module 603 trains the classification model according to the word vector matrix and the category label, specifically: inputting the word vector matrix and the category label into the neural network for iterative training, and inputting the neural force in each iteration training.
  • the word vector in the word vector matrix of the network and the parameters of the neural network are adjusted to generate a classification model.
  • FIG. 7 is a schematic structural diagram of an information output apparatus according to an embodiment of the present application.
  • the information output device 700 includes a processor 701, a memory 702, and a communication interface 703.
  • the processor 701, the memory 702 and the communication interface 703 are connected.
  • the processor 701 may be a central processing unit (CPU), a general-purpose processor, a coprocessor, a digital signal processor (DSP), or an application-specific integrated circuit (ASIC). , field programmable gate array (FPGA) or other programmable logic device, transistor logic device, hardware component, or any combination thereof.
  • the processor 701 can also be a combination of computing functions, such as one or more microprocessor combinations, a combination of a DSP and a microprocessor, and the like.
  • the communication interface 703 is used to implement communication with other network elements.
  • the processor 701 calls the program code stored in the memory 702 to execute the steps performed by the information output device in the above method embodiment.
  • FIG. 8 is a schematic structural diagram of a model training apparatus disclosed in an embodiment of the present application.
  • the model training device 800 includes a processor 801, a memory 802, and a communication interface 803.
  • the processor 801, the memory 802, and the communication interface 803 are connected.
  • the processor 801 can be a central processing unit (CPU), a general purpose processor, a coprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC). , field programmable gate array (FPGA) or other programmable logic device, transistor logic device, hardware component, or any combination thereof.
  • the processor 801 can also be a combination of computing functions, such as one or more microprocessor combinations, a combination of a DSP and a microprocessor, and the like.
  • the communication interface 803 is used to implement communication with other network elements.
  • the processor 801 calls the program code stored in the memory 802 to execute the steps performed by the model training device in the foregoing method embodiment.
  • each device provided in the embodiment of the present application is similar to the method embodiment of the present application. Therefore, the implementation of each device may refer to the implementation of the method, and is not described here.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Machine Translation (AREA)

Abstract

本申请公开了一种信息输出方法及装置,其中,该方法包括:获取故障描述文本,故障描述文本用于描述网络中发生的故障;通过语义生成模型生成故障描述文本的语义向量;获取多种类型的目标数据的相关文本分别对应的语义向量,该目标数据用于协助分析故障产生的原因;计算故障描述文本的语义向量与每种目标数据的相关文本的语义向量的相关性;确定并输出第一数据,该第一数据为多种目标数据中语义向量与故障描述文本的语义向量的相关性最大的目标数据,或该第一数据为多种目标数据中语义向量与故障描述文本的语义向量的相关性大于预设阈值的目标数据。通过实施本申请的方法,能够准确地查找出与故障描述文本的相关的用于协助分析故障原因的数据。

Description

一种信息输出方法及装置 技术领域
本申请涉及通信技术领域,尤其涉及一种信息输出方法及装置。
背景技术
当网络设备出现故障时,会影响正常的通信,给人们的工作和生活带来严重损失,所以网络设备故障的及时修复非常重要。目前,当网络设备出现故障时,一线工程师会从故障发生现场收集用于协助分析故障原因的数据,例如,收集网络设备故障发生前后一段时间内的关键性能指标(KPI)、设备告警、设备日志等参数数据。并且一线工程师会对故障现象进行描述,得到故障描述文本。一线工程师将收集的KPI等数据和故障描述文本以故障工单的形式反馈给运维部门。运维工程师根据故障工单中的故障描述文本,凭借自身的专业知识,手动从一线收集的数据中选择出一些KPI、设备告警、设备日志等参数数据。进一步地,对选择出来的这些数据进行异常检测和相互佐证,从而分析出故障根因所在,对故障网络设备的修复提供指导性意见。这种通过人工手动从KPI、设备告警、设备日志等参数数据中选择出与故障描述文本相关的参数数据进行查看分析的故障检测方法,效率低速度慢,无法满足日益增加的网络需求。
现有技术中通过查找与故障描述文本具有相同的关键词的文本,并根据该文本的相关参数数据进行故障的查看分析。但相关性高的能用于协助分析故障原因的相关文本和故障描述文本中可能并没有相同的关键词。因此,通过现有的方式不能准确地查找到与故障描述文本相关联的用于协助分析故障原因的数据。
发明内容
本申请提供了一种信息输出方法及装置,能够自动地并准确地查找到与故障描述文本的相关的用于协助分析故障原因的数据。
第一方面,本申请提供了一种信息输出方法,该方法包括:获取故障描述文本,该故障描述文本用于描述网络中发生的故障;通过语义生成模型生成故障描述文本的语义向量;获取多种类型的目标数据的相关文本分别对应的语义向量,该目标数据用于协助分析故障产生的原因;计算故障描述文本的语义向量与每种目标数据的相关文本的语义向量的相关性;确定并输出第一数据,该第一数据为每种目标数据中语义向量与故障描述文本的语义向量的相关性最大的目标数据,或该第一数据为每种目标数据中语义向量与故障描述文本的语义向量的相关性大于预设阈值的目标数据。
本申请通过对比故障描述文本的语义向量与目标数据的相关文本的语义向量的相关性,可以准确地查找到与故障描述文本相关联的目标数据。比如,故障描述为“行业用户上网慢”,本申请分析出的与其相关的用于故障分析的关键性指标的名称为“下行带宽控制丢包比例”。可以看出,从字面上二者没有任何可以匹配和关联的成分,而本申请恰恰是通过语义分析挖掘学习到了“上网速度和丢包比例有关系”这样的领域知识,才实现了二者相关联的分析。因此,通过实施第一方面所描述的方法,能够自动地并且准确地查找出与 故障描述文本的相关的用于协助分析故障原因的数据。
在一种可能的实施方式中,获取故障描述文本之前,还可通过语义生成模型生成多种类型的目标数据的相关文本分别对应的语义向量。
并且还可保存多种目标数据的相关文本分别对应的语义向量;相应地,获取多种目标数据的相关文本分别对应的语义向量的具体实施方式为:获取保存的多种目标数据的相关文本分别对应的语义向量。
通过实施该实施方式,可预先生成并保存多种目标数据的相关文本分别对应的语义向量,在接收故障描述文本之后,可直接用保存的多种目标数据的相关文本分别对应的语义向量与故障描述文本的语义向量进行相关性计算,从而不用在接收故障描述文本之后,临时生成多种目标数据的相关文本分别对应的语义向量。可见,通过实施该实施方式,有利于快速地计算得到故障描述文本的语义向量与每种目标数据的相关文本的语义向量的相关性。
在一种可能的实施方式中,上述语义生成模型是根据历史故障描述文本对应的词向量矩阵训练生成的,该词向量矩阵包括历史故障描述文本中各个词对应的词向量,该词向量用于表示词的语义。
通过实施该实施方式训练得到的语义生成模型,能够更加准确地表达文本的语义。
在一种可能的实施方式中,上述多种类型的目标数据包括关键性能指标、设备告警、设备日志中的至少两种;当上述目标数据为关键性能指标时,该目标数据的相关文本为关键性能指标的名称;当上述目标数据为设备告警时,该目标数据的相关文本为设备告警的标识;当上述目标数据为设备日志时,该目标数据的相关文本为设备日志的内容片段。
第二方面,本申请提供了一种语义生成模型的训练方法,该方法包括:获取训练文本对应的词向量集合,该词向量集合中包括的词向量与训练文本中的词一一对应,该词向量用于表示词的语义;根据词向量集合将历史故障描述文本转换为由至少一个词向量组成的词向量矩阵;根据词向量矩阵训练得到语义生成模型,该语义生成模型用于生成文本的语义向量。
可选的,获取训练文本对应的词向量集合之后,可将训练文本对应的词向量集合进行保存,以便后续使用词向量集合中的词向量。
可见,第二方面所描述的方法是从词汇层面的语义向句子层面的语义逐步建模得到语义生成模型,这种语义生成模型训练方式是符合语言生成的基本原理的。因此,通过实施第二方面所描述的方法训练得到的语义生成模型,能够更加准确地表达文本的语义。
在一种可能的实施方式中,根据词向量集合将历史故障描述文本转换为由至少一个词向量组成的词向量矩阵的具体实施方式为:对历史故障描述文本进行分词处理,得到历史故障描述文本对应的由至少一个词组成的词序列;从词向量集合中获取词序列包括的词对应的词向量;将词序列包括的各个词对应的词向量组成词向量矩阵。
通过实施该实施方式,可准确地将历史故障描述文本转换为由至少一个词向量组成的词向量矩阵。
在一种可能的实施方式中,当词向量集合中不存在词序列包括的词对应的词向量时,生成随机向量作为词序列包括的词对应的词向量。
通过实施该实施方式,可准确地将历史故障描述文本转换为由至少一个词向量组成的词向量矩阵。
在一种可能的实施方式中,根据词向量矩阵训练得到语义生成模型的具体实施方式为:获取历史故障描述文本对应的故障设备类型;根据词向量矩阵和类别标签训练分类模型,该类别标签包括该故障设备类型;根据分类模型得到语义生成模型。
通过实施该实施方式训练得到的语义生成模型,能够更加准确地表达文本的语义。
在一种可能的实施方式中,根据词向量矩阵和类别标签训练分类模型的具体实施方式为:将词向量矩阵和类别标签输入神经网络进行迭代训练,在每次迭代训练时对输入神经网络的词向量矩阵中的词向量和神经网络的参数进行调整,以生成该分类模型。通过该实施方式训练得到的语义生成模型,能够更加准确地表达文本的语义。
可选的,还可将使用最后一次迭代训练输入的词向量矩阵中的词向量更新词向量集合中相应词对应的词向量。通过实施该实施方式,能够根据带有领域知识的历史故障描述文本语料修正词向量集合中的词向量,使词向量集合中的词向量更能表达领域知识的词的语义信息。
第三方面,提供了一种信息输出装置,该信息输出装置可执行上述第一方面或第一方面可能的实施方式中的方法。该功能可以通过硬件实现,也可以通过硬件执行相应的软件实现。该硬件或软件包括一个或多个与上述功能相对应的单元。该单元可以是软件和/或硬件。基于同一发明构思,该信息输出装置解决问题的原理以及有益效果可以参见上述第一方面或第一方面可能的实施方式中以及有益效果,重复之处不再赘述。
第四方面,提供了一种模型训练装置,该模型训练装置可执行上述第二方面或第二方面可能的实施方式中的方法。该功能可以通过硬件实现,也可以通过硬件执行相应的软件实现。该硬件或软件包括一个或多个与上述功能相对应的单元。该单元可以是软件和/或硬件。基于同一发明构思,该模型训练装置解决问题的原理以及有益效果可以参见上述第二方面或第二方面可能的实施方式中以及有益效果,重复之处不再赘述。
第五方面,提供了一种信息输出装置,该信息输出装置包括:处理器、存储器、通信接口;处理器、通信接口和存储器相连;其中,通信接口可以为收发器。通信接口用于实现与其他网元之间的通信。其中,一个或多个程序被存储在存储器中,该处理器调用存储在该存储器中的程序以实现上述第一方面或第一方面可能的实施方式中的方案,该信息输出装置解决问题的实施方式以及有益效果可以参见上述第一方面或第一方面可能的实施方式以及有益效果,重复之处不再赘述。
第六方面,提供了一种模型训练装置,该模型训练装置包括:处理器、存储器、通信接口;处理器、通信接口和存储器相连;其中,通信接口可以为收发器。通信接口用于实现与其他网元之间的通信。其中,一个或多个程序被存储在存储器中,该处理器调用存储在该存储器中的程序以实现上述第二方面或第二方面可能的实施方式中的方案,该模型训练装置解决问题的实施方式以及有益效果可以参见上述第二方面或第二方面可能的实施方式以及有益效果,重复之处不再赘述。
第七方面,提供了一种计算机程序产品,当其在计算机上运行时,使得计算机执行上述第一方面、第二方面、第一方面的可能的实施方式或第二方面的可能的实施方式中的方 法。
第八方面,提供了一种信息输出装置的芯片产品,执行上述第一方面或第一方面的任意可能的实施方式中的方法。
第九方面,提供了一种模型训练装置的芯片产品,执行上述第二方面或第二方面的任意可能的实施方式中的方法。
第十方面,提了供一种计算机可读存储介质,计算机可读存储介质中存储有指令,当其在计算机上运行时,使得计算机执行上述第一方面的方法或第一方面的可能的实施方式中的方法。
第十一方面,提了供一种计算机可读存储介质,计算机可读存储介质中存储有指令,当其在计算机上运行时,使得计算机执行上述第二方面的方法或第二方面的可能的实施方式中的方法。
附图说明
图1是本申请实施例提供的一种信息输出方法的流程示意图;
图2是本申请实施例提供的一种语义生成模型的训练方法的流程示意图;
图3是本申请实施例提供的一种CBOW算法采用的神经网络的示意图;
图4是本申请实施例提供的一种用于训练分类模型的神经网络的结构示意图;
图5是本申请实施例提供的一种信息输出装置的结构示意图;
图6是本申请实施例提供的一种模型训练装置的结构示意图;
图7是本申请实施例提供的另一种信息输出装置的结构示意图;
图8是本申请实施例提供的另一种模型训练装置的结构示意图。
具体实施方式
下面结合附图对本申请具体实施例作进一步的详细描述。
本申请实施例提供了一种信息输出方法及装置,能够自动确定并输出与故障描述文本的相关的用于协助分析故障原因的数据。
以下对本申请所提供的信息输出方法及装置进行详细地介绍。
请参见图1,图1是本申请实施例提供的一种信息输出方法的流程示意图。如图1所示,该信息输出方法包括如下101~105部分,其中:
101、信息输出装置获取故障描述文本。
其中,故障描述文本为对故障现象进行描述的文本,即故障描述文本用于描述网络中发生的故障。例如,故障描述文本可以为“行业用户上网慢”、“在线计费服务器(online charging system,OCS)通讯中断”等。故障描述文本可以是其他装置发送给信息输出装置的。例如,一线工程师对故障现象进行描述,得到故障描述文本,并将收集的用于协助分析故障原因的数据(如关键性能指标等)和故障描述文本以故障工单的形式发送给运维部门的信息输出装置。
102、信息输出装置通过语义生成模型生成故障描述文本的语义向量。
在一种可能的实施方式中,语义生成模型可以是根据历史故障描述文本对应的词向量 矩阵训练生成的,该词向量矩阵包括历史故障描述文本中各个词对应的词向量。
可选的,语义生成模型的训练方法具体可参见下图2所描述的语义生成模型的训练方法。也就是说,信息输出装置所使用的语义生成模型可以是下图2中模型训练装置训练的语义生成模型。图1中的信息输出装置和图2中的模型训练装置可部署在同一设备或部署在不同的设备。当图1中的信息输出装置和图2中的模型训练装置部署在不同的设备中时,模型训练装置训练完语义生成模型之后可发送语义生成模型至信息输出装置,从而信息输出装置可通过接收的语义生成模型生成故障描述文本的语义向量。当图1中的信息输出装置和图2中的模型训练装置部署在相同的设备中时,信息输出装置可从模型训练装置中获取语义生成模型,从而信息输出装置可通过语义生成模型生成故障描述文本的语义向量。
当然语义生成模型也可以不通过图2所描述的方式训练生成,也可通过其他方式训练生成,本申请实施例不做限定。
在一种可能的实施方式中,信息输出装置通过语义生成模型生成故障描述文本的语义向量的具体实施方式为:
信息输出装置根据词向量集合将故障描述文本转换为词向量矩阵,再将词向量矩阵输入语义生成模型,以生成故障描述文本的语义向量,其中,该词向量集合中包括多个词向量。可选的,该词向量集合可以是下图2中的模型训练装置生成并发送至信息输出装置的。
可选的,信息输出装置根据词向量集合将故障描述文本转换为词向量矩阵的具体实施方式为:信息输出装置对故障描述文本进行分词处理,得到故障描述文本对应的由至少一个词组成的词序列;从词向量集合中获取该词序列包括的词对应的词向量;将该词序列包括的各个词对应的词向量组成故障描述文本的词向量矩阵。当词向量集合中不存在该词序列包括的词对应的词向量时,生成随机向量作为该词序列包括的词对应的词向量。
举例来说,故障描述文本包括4个词,对故障描述文本进行分词处理得到的词序列为“行业”、“用户”、“上网”、“慢”。信息输出装置从词向量集合中查找到“行业”对应词向量1、“用户”对应词向量2、“上网”对应词向量3,未查找到“慢”对应的词向量,则生成随机向量词向量4作为“慢”对应的词向量。信息输出装置将词向量1~4组成故障描述文本的词向量矩阵。再将该词向量矩阵输入语义生成模型中生成故障描述文本的语义向量。
103、信息输出装置获取多种类型的目标数据的相关文本分别对应的语义向量。
其中,该目标数据用于协助分析故障产生的原因。其中,103部分和102部分的执行顺序可以不分先后,可先执行102部分再执行103部分,或可先执行103部分再执行102部分。
在一种可能的实施方式中,该多种类型的目标数据包括关键性能指标(KPI)、设备告警、设备日志中的至少两种;当目标数据为关键性能指标时,该目标数据的相关文本为关键性能指标的名称;当目标数据为设备告警时,该目标数据的相关文本为设备告警的标识;当目标数据为设备日志时,该目标数据的相关文本为设备日志的内容片段。其中,每种类型的目标数据为多个。
例如,该多种类型的目标数据包括关键性能指标和设备告警。上述的多种类型的目标数据为100个不同的关键性能指标和20个不同的设备告警,100个关键性能指标分别为关键性能指标1~关键性能指标100。20个设备告警分别为设备告警1~20。信息输出装置获取 的多种类型的目标数据的相关文本分别对应的语义向量为关键性能指标1~关键性能指标100的名称分别对应的语义向量,以及设备告警1~20的标识分别对应的语义向量。也就是说,信息输出装置会获取120个语义向量。
在一种可能的实施方式中,信息输出装置可在接收故障描述文本之前,通过语义生成模型生成多种类型的目标数据的相关文本分别对应的语义向量。
可选的,信息输出装置生成多种类型的目标数据的相关文本分别对应的语义向量之后,可保存该多种类型的目标数据的相关文本分别对应的语义向量。在接收故障描述文本之后,就可获取保存该多种目标数据的相关文本分别对应的语义向量,以便与故障描述文本的语义向量进行相关性计算。通过实施该实施方式,可预先生成并保存多种目标数据的相关文本分别对应的语义向量,在接收故障描述文本之后,可直接用保存的多种目标数据的相关文本分别对应的语义向量与故障描述文本的语义向量进行相关性计算,从而不用在接收故障描述文本之后,临时生成多种目标数据的相关文本分别对应的语义向量。可见,通过实施该实施方式,有利于快速地计算得到故障描述文本的语义向量与每种目标数据的相关文本的语义向量的相关性。
在一种可能的实施方式中,信息输出装置通过语义生成模型生成目标数据的相关文本对应的语义向量的原理与信息输出装置通过语义生成模型生成故障描述文本的语义向量的原理相同,在此不赘述。
104、信息输出装置计算故障描述文本的语义向量与每种目标数据的相关文本的语义向量的相关性。
举例来说,具有两种类型的目标数据,分别为100个不同的关键性能指标和20个不同的设备告警,100个关键性能指标分别为关键性能指标1~关键性能指标100。20个设备告警分别为设备告警1~20。信息输出装置计算故障描述文本的语义向量分别与100个关键性能指标的相关文本的语义向量的相关性,以及计算故障描述文本的语义向量分别与20个设备告警的相关文本的语义向量的相关性。因此,会得到120个相关性。
在一种可能的实施方式中,可采用向量的夹角来作为相关性的衡量,故障描述文本的语义向量与目标数据的相关文本的语义向量的相关性可表示为:
Figure PCTCN2019084814-appb-000001
其中,cos(θ)为故障描述文本的语义向量与目标数据的相关文本的语义向量的相关性,n为故障描述文本和目标数据的相关文本的语义向量的维度数量,x i为故障描述文本第i维的语义向量,y i目标数据的相关文本第i维的语义向量。
105、信息输出装置确定并输出第一数据。
其中,在信息输出装置计算故障描述文本的语义向量与每种目标数据中的每种目标数据的相关文本的语义向量的相关性之后,信息输出装置确定并输出第一数据,该第一数据为该每种目标数据中语义向量与故障描述文本的语义向量相关性最大的目标数据,或该第一数据为该每种目标数据中语义向量与故障描述文本的语义向量相关性大于预设阈值的目标数据。
例如,获取的两种类型的目标数据,分别为100个不同的关键性能指标和20个不同的设备告警,100个关键性能指标分别为关键性能指标1~关键性能指标100。20个设备告警分别为设备告警1~设备告警20。故障描述文本的语义向量与关键性能指标1~关键性能指标100的相关文本的语义向量的相关性为分别为相关性1~100。相关性1为最大的相关性,则信息输出装置输出关键性能指标1。故障描述文本的语义向量与设备告警1~设备告警20的相关文本的语义向量的相关性分别为相关性101~120。相关性120为最大的相关性,则信息输出装置输出设备告警20。
再如,相关性1和相关性2为大于预设阈值的相关性,则信息输出装置输出关键性能指标1和关键性能指标2。相关性101和相关性102为大于预设阈值的相关性,则信息输出装置输出设备告警1和设备告警2。
语义向量与故障描述文本的语义向量相关性越大的目标数据,说明该目标数据与故障描述文本越相关,用户可能需要查看该目标数据以分析故障原因。例如,故障描述文本为“ocs通讯中断”,关键性指标的名称为“ocs通讯中断次数”,该故障描述文本的语义向量与关键性指标的名称的语义向量相关性很大,用户可能需要查看该关键性指标以分析故障原因。可见,通过实施图1所描述的方法,能够自动查找到与故障描述文本的相关的用于协助分析故障原因的数据。
现有技术中通过查找与故障描述文本具有相同的关键词的文本,并根据该文本的相关参数数据进行故障的查看分析。但相关性高的能用于协助分析故障原因的相关文本和故障描述文本中可能并没有相同的关键词。因此,通过现有的方式不能准确地查找到与故障描述文本相关联的用于协助分析故障原因的数据。本申请实施例通过对比故障描述文本的语义向量与目标数据的相关文本的语义向量的相关性,可以准确地查找到与故障描述文本相关联的目标数据。比如,故障描述为“行业用户上网慢”,本申请实施例分析出的与其相关的用于故障分析的关键性指标的名称为“下行带宽控制丢包比例”。可以看出,从字面上二者没有任何可以匹配和关联的成分,而本申请恰恰是通过语义分析挖掘学习到了“上网速度和丢包比例有关系”这样的领域知识,才实现了二者相关联的分析。
因此,通过实施图1所描述的方法,能够自动地并且准确地查找到与故障描述文本的相关的用于协助分析故障原因的数据。
请参见图2,图2是本申请实施例提供的一种语义生成模型的训练方法的流程示意图。如图2所示,该语义生成模型的训练方法包括如下201~203部分,其中:
201、模型训练装置获取训练文本对应的词向量集合。
其中,词向量集合中包括的词向量与训练文本中的词一一对应。例如,训练文本中包括10000个词,则词向量集合中也包括10000个词向量。该词向量用于表示词的语义。可选的,获取训练文本对应的词向量集合之后,可保存训练文本对应的词向量集合,以便后续使用词向量集合中的词向量。
训练文本即语料。在一种可能的实施方式中,训练文本可以为百科类文本。从百科类文本中学习得到的词向量具有很好的通用语义。
在一种可能的实施方式中,模型训练装置首先对训练文本进行预处理,按句切分后再 对每句文本进行分词处理,得到分词后的训练文本,并通过word2vec工具或其他工具获取分词后的训练文本对应的词向量集合。
例如,训练文本为“数学是利用符号语言研究数量结构变化以及空间等概念的一门学科。我喜欢数学”。模型训练装置向将训练文本拆分为两句话,分别为“数学是利用符号语言研究数量结构变化以及空间等概念的一门学科”和“我喜欢数学”。再对这两个句子分别进行分词处理。得到分词后的训练文本为“数学是利用符号语言研究数量结构变化以及空间等概念的一门学科。我喜欢数学”。模型训练装置使用word2vec工具对分词后的训练文本进行逐句遍历,遍历结束就得到了训练文本中的每个词对应的词向量。模型训练装置将训练文本中的每个词对应的词向量组成的词向量集合进行保存。
模型训练装置可通过word2vec工具并采用CBOW算法获取分词后的训练文本对应的词向量集合。CBOW算法的思想是通过给定的上下文词来预测当前词。CBOW算法训练的目标是给定某个词的上下文时,使得该词出现的概率最大。训练结束后,每个词在输出层都得到了一个对应的词向量。尽管CBOW算法的建模思想是一个分类过程,但会生成词向量这一副产品。
例如,图3为CBOW算法采用的神经网络的示意图。如图3所示,该神经网络由三层结构构成,分别为输入层、映射层和输出层。其中,输出层包括已经构造好的哈夫曼树。哈夫曼树的一个叶子节点代表训练文本中的一个词的词向量,每个叶子节点对应的单词的词向量是随机初始化的。每个非叶节点内置一个权重向量,该向量的维度和输入层的词向量相同。
其中,输入层为某个单词w(t)周围的n-1个单词的词向量。n为窗口大小。例如,如果n取5,单词w(t)周围的n-1个单词为单词w(t)的前两个和后两个的单词。单词w(t)的前两个和后两个的单词分别为w(t-2)、w(t-1)、w(t+1)、w(t+2)。相对应的,这n-1个单词的词向量记为v(w(t-2))、v(w(t-1))、v(w(t+1))、v(w(t+2))。输入层将到这n-1个词向量传递到映射层,映射层将n-1个词向量进行相加,即将n-1个词向量的各维度对应相加。例如,映射层输入为pro(t)=v(w(t-2))+v(w(t-1))+v(w(t+1))+v(w(t+2))。
投影层将加和得到的向量pro(t)输入到哈夫曼树的根节点。在将pro(t)输入根节点之后,会计算根节点到每个叶子节点的概率,模型的训练过程是期望能得到由根节点到达w(t)对应的叶子节点的概率最大,由于在海量的训练文本中,相同的上下文环境会多次出现,所以在遍历训练训练文本的过程中会不断修正各权重向量,达到这样的效果。对训练文本中的所有词遍历完成之后,哈夫曼树的各叶子节点对应的词向量就为训练文本的各个词对应的词向量。这里的“训练文本中的所有词”包括训练文本中重复的词。
其中,从根节点到词w(t)对应的叶子节点每次经过一个中间节点时,相当于是进行了一次二分类,分类器可以采用softmax回归分类器。其中,每次分类的分类概率为:
Figure PCTCN2019084814-appb-000002
其中,θ i表示第i个权重向量。pro(t)为w(t)的上下文的词向量之和,e为自然常数。
设由根节点遍历到词w(t)对应的叶子节点的路径包含了L个中间节点,这些节点上的 参数组成参数向量为[θ 1,θ 2,θ 3,…,θ L],则根节点到词w(t)对应的叶子节点的概率为每次二分类的概率的乘积,即根节点到词w(t)对应的叶子节点的概率为:
Figure PCTCN2019084814-appb-000003
其中,P(w(t)|context(w(t)))为根节点到词w(t)对应的叶子节点的概率,
Figure PCTCN2019084814-appb-000004
表示i从1到L逐一递增对P(context(w(t)),θ i)连乘求积。从根节点到其他叶子节点的概率的计算方法同理,在此不赘述。
202、模型训练装置根据词向量集合将历史故障描述文本转换为由至少一个词向量组成的词向量矩阵。
具体地,模型训练装置可将大量的历史故障描述文本转换为词向量矩阵。模型训练装置根据大量的词向量矩阵训练得到语义生成模型。例如,具有历史故障描述文本1~历史故障描述文本100,可将历史故障描述文本1~历史故障描述文本100分别转换为词向量矩阵,即得到100个词向量矩阵。模型训练装置根据这100个词向量矩阵训练得到语义生成模型。
在一种可能的实施方式中,模型训练装置根据词向量集合将历史故障描述文本转换为由至少一个词向量组成的词向量矩阵的具体实施方式可以为:模型训练装置对历史故障描述文本进行分词处理,得到历史故障描述文本对应的由至少一个词组成的词序列;从词向量集合中获取该词序列包括的词对应的词向量;将该词序列包括的各个词对应的词向量组成该历史故障描述文本的词向量矩阵。当词向量集合中不存在该词序列包括的词对应的词向量时,可生成随机向量作为该词序列包括的词对应的词向量。可见,通过实施该实施方式,可准确地将历史故障描述文本转换为由至少一个词向量组成的词向量矩阵。
举例来说,历史故障描述文本1包括4个词,对历史故障描述文本1进行分词处理得到的词序列为“行业”、“用户”、“上网”、“慢”。模型训练装置从词向量集合中查找到“行业”对应词向量1、“用户”对应词向量2、“上网”对应词向量3,未查找到“慢”对应的词向量,则生成随机向量词向量4作为“慢”对应的词向量。模型训练装置将词向量1~4组成该历史故障描述文本1的词向量矩阵1。其他历史故障描述文本2~100转换为词向量矩阵的原理与历史故障描述文本1转换为词向量矩阵的原理相同,在此不赘述。
203、模型训练装置根据词向量矩阵训练得到语义生成模型。
具体地,模型训练装置在得到词向量矩阵之后,可将词向量矩阵输入神经网络进行训练,以得到语义生成模型。该语义生成模型用于生成文本的语义向量。该语义向量用于表示文本的语义。
可见,图2所描述的方法是从词汇层面的语义向句子层面的语义逐步建模得到语义生成模型,这种语义生成模型训练方式是符合语言生成的基本原理的。因此,通过实施图2所描述的方法训练得到的语义生成模型,能够更加准确地表达文本的语义。
在一种可能的实施方式中,模型训练装置根据词向量矩阵训练得到语义生成模型的具体实施方式为:模型训练装置获取历史故障描述文本对应的故障设备类型;模型训练装置根据词向量矩阵和类别标签训练分类模型,该类别标签包括该故障设备类型;模型训练装置根据分类模型得到语义生成模型。通过实施该实施方式训练得到的语义生成模型,能够 更加准确地表达文本的语义。
例如,历史故障描述文本对应的故障设备类型可以为路由器、有线设备或无线设备等。例如,历史故障描述文本描述的故障为路由器产生的故障,则历史故障描述文本对应的故障设备类型为路由器。一线工程师可收集每个故障描述文本对应的故障设备类型,然后将故障描述文本、故障描述文本对应的故障设备类型和用于协助分析故障原因的数据添加至工单中,并将工单发送给运维终端进行故障原因分析。因此,模型训练装置可从工单中获取历史故障描述文本对应的故障设备类型。
其中,训练得到的分类模型是用于生成故障描述文本对应的故障设备类型的模型。例如,将故障描述文本1对应的词向量矩阵输入分类模型,该分类模型可输出故障描述文本1对应的故障设备类型。
在一种可能的实施方式中,模型训练装置根据词向量矩阵和类别标签训练分类模型的具体实施方式为:将词向量矩阵和类别标签输入神经网络进行迭代训练,在每次迭代训练时对输入神经网络的词向量矩阵中的词向量和神经网络的参数进行调整,以生成分类模型。通过实施该实施方式,能够使训练得到的分类模型能够准确地对故障描述文本进行分类。
可选的,模型训练装置还可使用调整后的词向量矩阵中的词向量更新词向量集合中相应词对应的词向量。通过实施该可选的方式,能够根据带有领域知识的历史故障描述文本语料修正词向量集合中的词向量,使词向量集合中的词向量更能表达故障领域的词的语义信息。
举例来说,图4为一种用于训练分类模型的神经网络的结构示意图。如图4所示,该神经网络包括卷积层、池化层和全连接层。历史故障描述文本1的词向量矩阵1包括词向量{w1,w2,w3,w4,w5,w6}。每个词向量的维度为128个维度。模型训练装置得到词向量矩阵1之后,将词向量矩阵1输入神经网络。如图4所示,神经网络中的具有两个卷积核。当然在实际应用中也可以有两个以上的卷积核,本申请实施例以两个卷积核进行举例说明。左边的卷积核1对词向量矩阵1包括词向量进行两两卷积。例如,w1与w2进行卷积得到C1,w2与w3进行卷积得到C2,w3与w4进行卷积得到C3,w4与w5进行卷积得到C4,w5与w6进行卷积得到C5。右边的卷积核2对词向量矩阵1包括词向量进行三三卷积。例如,w1、w2和w3进行卷积得到C6,w2、w3和w4进行卷积得到C7,w3、w4和w5进行卷积得到C8,w4、w5和w6进行卷积得到C9。实际应用中也对其他数量的词向量进行卷积,本申请实施例以两两卷积和三三卷积进行举例说明。
可见,卷积核1可生成一个特征图(feature map)C=[C1,C2,…,C5],卷积核2生成一个特征图C=[C6,C7,C8,C9]。模型训练装置得到每个卷积核生成的特征图之后,针对每个特征图,通过最大池化操作选取每个维度上的最大值作为当前卷积核生成的文本特征向量。模型训练装置将所有文本特征向量进行拼接,得到最终的历史故障描述文本1的语义向量。即如图4所示,模型训练装置从C1~C5的第一个维度中选取最大的值,从C1~C5的第2个维度中选取最大的值,从C1~C5的第3个维度中选取最大的值,依次类推,直到从C1~C5的第128个维度中选取到最大的值。模型训练装置将选取的128个维度的最大值组成卷积核1对应的文本特征向量1。同理,模型训练装置也获取卷积核2对应的文本特征向量2。模型训练装置将文本特征向量1和文本特征向量2进行拼接,得到最终的 历史故障描述文本1的语义向量。
模型训练装置将得到的历史故障描述文本1的语义向量输入全连接层,并将历史故障描述文本1对应的故障设备类型(如路由器)作为类别标签,输入全连接层。模型训练装置在全连接层对历史故障描述文本1的语义向量进行分析,分析得到故障设备类型最大概率为交换机。由于历史故障描述文本1的语义向量进行分析得到的最大概率的故障设备类型(即交换机)与历史故障描述文本1对应的类别标签(即路由器)不相同,因此模型训练装置记录通过对历史故障描述文本1的语义向量进行分析得到的最大概率的故障设备类型不正确。同理,模型训练装置按照上述流程将历史故障描述文本2的词向量矩阵输入神经网络进行训练,得到历史故障描述文本2的语义向量,并在全连接层输入历史故障描述文本2对应的故障设备类型(如交换机)作为类别标签。模型训练装置对历史故障描述文本2的语义向量进行分析,分析得到故障设备类型最大概率为防火墙。因此,模型训练装置记录通过对历史故障描述文本2的语义向量进行分析得到的最大概率的故障设备类型不正确。假设具有100个历史故障描述文本,其余98个历史故障描述文本同理,均按照上述历史故障描述文本1的方式输入神经网络进行分类模型的训练。在对历史故障描述文本1~100完成第一轮训练之后,假设根据历史故障描述文本1~50对应的语义向量分析得到的最大概率的故障设备类型不正确,模型训练装置对神经网络的参数以及历史故障描述文本1~50对应的词向量矩阵中的词向量进行调整。调整完毕之后,以新词向量矩阵和神经网络的参数重新对历史故障描述文本1~100进行训练,直到根据历史故障描述文本1~100对应的语义向量分析得到的最大概率的故障设备类型与分类标签相匹配,就生成分类模型,即通过迭代训练神经网络来生成分类模型。
最后模型训练装置使用最后一轮迭代训练输入的词向量矩阵中的词向量更新词向量集合中的相应词对应的词向量。例如,历史故障描述文本1为“上网速度慢”最后一轮迭代训练之前对历史故障描述文本1对应词向量矩阵进行了调整,将“上网”对应的词向量调整为词向量1,则最后一次迭代训练完成后,使用词向量1替换词向量集合中的“上网”对应的词向量。历史故障描述文本2为“OCS通讯中断”最后一轮迭代训练之前对历史故障描述文本2对应词向量矩阵进行了调整,将“中断”对应的词向量调整为词向量2,则最后一轮迭代训练完成后,使用词向量2替换词向量集合中的“中断”对应的词向量。其他历史故障描述文本同理,在此不赘述。
在一种可能的实施方式中,模型训练装置根据分类模型得到语义生成模型的具体实施方式为:模型训练装置将分类模型中全连接层以上的部分作为语义生成模型。通过实施该实施方式生成的语义生成模型,可以准确的生成文本的语义向量。
本发明实施例可以根据上述方法示例对设备进行功能模块的划分,例如,可以对应各个功能划分各个功能模块,也可以将两个或两个以上的功能集成在一个模块中。上述集成的模块既可以采用硬件的形式实现,也可以采用软件功能模块的形式实现。需要说明的是,本发明实施例中对模块的划分是示意性的,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式。
请参见图5,图5是本发明实施提供的一种信息输出装置。该信息输出装置包括:获 取模块501、生成模块502、计算模块503和输出模块504。其中:
获取模块501,用于获取故障描述文本;生成模块502,用于通过语义生成模型生成故障描述文本的语义向量,该故障描述文本用于描述网络中发生的故障;获取模块501,还用于获取多种类型的目标数据的相关文本分别对应的语义向量,该目标数据用于协助分析故障产生的原因;计算模块503,用于计算故障描述文本的语义向量与每种目标数据的相关文本的语义向量的相关性;输出模块504,用于确定并输出第一数据,第一数据为每种目标数据中语义向量与故障描述文本的语义向量的相关性最大的目标数据,或第一数据为每种目标数据中语义向量与故障描述文本的语义向量的相关性大于预设阈值的目标数据。
在一种可能的实施方式中,生成模块502,还用于在获取模块501获取故障描述文本之前,通过语义生成模型生成多种目标数据的相关文本分别对应的语义向量。
在一种可能的实施方式中,语义生成模型是根据历史故障描述文本对应的词向量矩阵训练生成的,词向量矩阵包括历史故障描述文本中各个词对应的词向量,该词向量用于表示词的语义。
在一种可能的实施方式中,该多种类型的目标数据包括关键性能指标、设备告警、设备日志中的至少两种;当目标数据为关键性能指标时,目标数据的相关文本为关键性能指标的名称;当目标数据为设备告警时,目标数据的相关文本为设备告警的标识;当目标数据为设备日志时,目标数据的相关文本为设备日志的内容片段。
请参见图6,图6是本发明实施提供的一种模型训练装置。该模型训练装置包括获取模块601、转换模块602和训练模块603,其中:
获取模块601,用于获取训练文本对应的词向量集合,词向量集合中包括的词向量与训练文本中的词一一对应;转换模块602,用于根据词向量集合将历史故障描述文本转换为由至少一个词向量组成的词向量矩阵;训练模块603,还用于根据词向量矩阵训练得到语义生成模型,语义生成模型用于生成文本的语义向量。
在一种可能的实施方式中,转换模块602具体用于:对历史故障描述文本进行分词处理,得到历史故障描述文本对应的由至少一个词组成的词序列;从词向量集合中获取词序列包括的词对应的词向量;将词序列包括的各个词对应的词向量组成词向量矩阵。
在一种可能的实施方式中,转换模块602还具体用于:当词向量集合中不存在词序列包括的词对应的词向量时,生成随机向量作为词序列包括的词对应的词向量。
在一种可能的实施方式中,训练模块603根据词向量矩阵训练得到语义生成模型的方式具体为:获取历史故障描述文本对应的故障设备类型;根据词向量矩阵和类别标签训练分类模型,该类别标签包括所述故障设备类型;根据分类模型得到语义生成模型。
在一种可能的实施方式中,训练模块603根据词向量矩阵和类别标签训练分类模型的方式具体为:将词向量矩阵和类别标签输入神经网络进行迭代训练,在每次迭代训练时对输入神经网络的词向量矩阵中的词向量和神经网络的参数进行调整,以生成分类模型。
请参见图7,图7是本申请实施例公开的一种信息输出装置的结构示意图。如图7所示,该信息输出装置700包括处理器701、存储器702和通信接口703。其中,处理器701、 存储器702和通信接口703相连。
其中,处理器701可以是中央处理器(central processing unit,CPU),通用处理器,协处理器,数字信号处理器(digital signal processor,DSP),专用集成电路(application-specific integrated circuit,ASIC),现场可编程门阵列(field programmable gate array,FPGA)或者其他可编程逻辑器件、晶体管逻辑器件、硬件部件或者其任意组合。该处理器701也可以是实现计算功能的组合,例如包含一个或多个微处理器组合,DSP和微处理器的组合等等。
其中,通信接口703用于实现与其他网元之间的通信。
其中,处理器701调用存储器702中存储的程序代码,可执行上述方法实施例中信息输出装置所执行的步骤。
请参见图8,图8是本申请实施例公开的一种模型训练装置的结构示意图。如图8所示,该模型训练装置800包括处理器801、存储器802和通信接口803。其中,处理器801、存储器802和通信接口803相连。
其中,处理器801可以是中央处理器(central processing unit,CPU),通用处理器,协处理器,数字信号处理器(digital signal processor,DSP),专用集成电路(application-specific integrated circuit,ASIC),现场可编程门阵列(field programmable gate array,FPGA)或者其他可编程逻辑器件、晶体管逻辑器件、硬件部件或者其任意组合。该处理器801也可以是实现计算功能的组合,例如包含一个或多个微处理器组合,DSP和微处理器的组合等等。
其中,通信接口803用于实现与其他网元之间的通信。
其中,处理器801调用存储器802中存储的程序代码,可执行上述方法实施例中模型训练装置所执行的步骤。
基于同一发明构思,本申请实施例中提供的各设备解决问题的原理与本申请方法实施例相似,因此各设备的实施可以参见方法的实施,为简洁描述,在这里不再赘述。
在上述实施例中,对各个实施例的描述都各有侧重,某个实施例中没有详述的部分,可以参见其他实施例的相关描述。
最后应说明的是:以上各实施例仅用以说明本申请的技术方案,而非对其限制;尽管参照前述各实施例对本申请进行了详细的说明,本领域的普通技术人员应当理解:其依然可以对前述各实施例所记载的技术方案进行修改,或者对其中部分或者全部技术特征进行等同替换;而这些修改或者替换,并不使相应技术方案的本质脱离本申请各实施例技术方案的范围。

Claims (20)

  1. 一种信息输出方法,其特征在于,所述方法包括:
    获取故障描述文本,所述故障描述文本用于描述网络中发生的故障;
    通过语义生成模型生成所述故障描述文本的语义向量;
    获取多种类型的目标数据的相关文本分别对应的语义向量,所述目标数据用于协助分析所述故障产生的原因;
    计算所述故障描述文本的语义向量与每种所述目标数据的相关文本的语义向量的相关性;
    确定并输出第一数据,所述第一数据为每种所述目标数据中语义向量与所述故障描述文本的语义向量的相关性最大的目标数据,或所述第一数据为每种所述目标数据中语义向量与所述故障描述文本的语义向量的相关性大于预设阈值的目标数据。
  2. 根据权利要求1所述的方法,其特征在于,所述获取故障描述文本之前,所述方法还包括:
    通过所述语义生成模型生成多种类型的目标数据的相关文本分别对应的语义向量。
  3. 根据权利要求1或2所述的方法,其特征在于,所述语义生成模型是根据历史故障描述文本对应的词向量矩阵训练生成的,所述词向量矩阵包括所述历史故障描述文本中各个词对应的词向量,所述词向量用于表示词的语义。
  4. 根据权利要求1~3任意一项所述的方法,其特征在于,所述多种类型的目标数据包括关键性能指标、设备告警、设备日志中的至少两种;当所述目标数据为所述关键性能指标时,所述目标数据的相关文本为所述关键性能指标的名称;当所述目标数据为所述设备告警时,所述目标数据的相关文本为所述设备告警的标识;当所述目标数据为所述设备日志时,所述目标数据的相关文本为所述设备日志的内容片段。
  5. 一种语义生成模型的训练方法,其特征在于,所述方法包括:
    获取训练文本对应的词向量集合,所述词向量集合中包括的词向量与所述训练文本中的词一一对应,所述词向量用于表示词的语义;
    根据所述词向量集合将历史故障描述文本转换为由至少一个词向量组成的词向量矩阵;
    根据所述词向量矩阵训练得到语义生成模型,所述语义生成模型用于生成文本的语义向量。
  6. 根据权利要求5所述的方法,其特征在于,所述根据所述词向量集合将历史故障描述文本转换为由至少一个词向量组成的词向量矩阵,包括:
    对历史故障描述文本进行分词处理,得到所述历史故障描述文本对应的由至少一个词组成的词序列;
    从所述词向量集合中获取所述词序列包括的词对应的词向量;
    将所述词序列包括的各个词对应的词向量组成词向量矩阵。
  7. 根据权利要求6所述的方法,其特征在于,所述方法还包括:
    当所述词向量集合中不存在所述词序列包括的词对应的词向量时,生成随机向量作为所述词序列包括的词对应的词向量。
  8. 根据权利要求5~7任意一项所述的方法,其特征在于,所述根据所述词向量矩阵训练得到语义生成模型,包括:
    获取所述历史故障描述文本对应的故障设备类型;
    根据所述词向量矩阵和类别标签训练分类模型,所述类别标签包括所述故障设备类型;
    根据所述分类模型得到语义生成模型。
  9. 根据权利要求8所述的方法,其特征在于,所述根据所述词向量矩阵和所述类别标签训练分类模型,包括:
    将所述词向量矩阵和所述类别标签输入神经网络进行迭代训练,在每次迭代训练时对输入所述神经网络的词向量矩阵中的词向量和所述神经网络的参数进行调整,以生成所述分类模型。
  10. 一种信息输出装置,其特征在于,所述信息输出装置包括:
    获取模块,用于获取故障描述文本,所述故障描述文本用于描述网络中发生的故障;
    生成模块,用于通过语义生成模型生成所述故障描述文本的语义向量;
    所述获取模块,还用于获取多种类型的目标数据的相关文本分别对应的语义向量,所述目标数据用于协助分析所述故障产生的原因;
    计算模块,用于计算所述故障描述文本的语义向量与每种所述目标数据的相关文本的语义向量的相关性;
    输出模块,用于确定并输出第一数据,所述第一数据为每种所述目标数据中语义向量与所述故障描述文本的语义向量的相关性最大的目标数据,或所述第一数据为每种所述目标数据中语义向量与所述故障描述文本的语义向量的相关性大于预设阈值的目标数据。
  11. 根据权利要求10所述的装置,其特征在于,
    所述生成模块,还用于在所述获取模块获取故障描述文本之前,通过所述语义生成模型生成多种目标数据的相关文本分别对应的语义向量。
  12. 根据权利要求10或11所述的装置,其特征在于,所述语义生成模型是根据历史故障描述文本对应的词向量矩阵训练生成的,所述词向量矩阵包括所述历史故障描述文本中各个词对应的词向量,所述词向量用于表示词的语义。
  13. 根据权利要求10~12任意一项所述的装置,其特征在于,所述多种类型的目标数据包括关键性能指标、设备告警、设备日志中的至少两种;当所述目标数据为所述关键性能指标时,所述目标数据的相关文本为所述关键性能指标的名称;当所述目标数据为所述设备告警时,所述目标数据的相关文本为所述设备告警的标识;当所述目标数据为所述设备日志时,所述目标数据的相关文本为所述设备日志的内容片段。
  14. 一种模型训练装置,其特征在于,所述模型训练装置包括:
    获取模块,用于获取训练文本对应的词向量集合,所述词向量集合中包括的词向量与所述训练文本中的词一一对应;
    转换模块,用于根据所述词向量集合将历史故障描述文本转换为由至少一个词向量组成的词向量矩阵;
    所述训练模块,还用于根据所述词向量矩阵训练得到语义生成模型,所述语义生成模型用于生成文本的语义向量。
  15. 根据权利要求14所述的装置,其特征在于,所述转换模块具体用于:
    对历史故障描述文本进行分词处理,得到所述历史故障描述文本对应的由至少一个词组成的词序列;
    从所述词向量集合中获取所述词序列包括的词对应的词向量;
    将所述词序列包括的各个词对应的词向量组成词向量矩阵。
  16. 根据权利要求15所述的装置,其特征在于,所述转换模块还具体用于:
    当所述词向量集合中不存在所述词序列包括的词对应的词向量时,生成随机向量作为所述词序列包括的词对应的词向量。
  17. 根据权利要求14~16任意一项所述的装置,其特征在于,所述训练模块根据所述词向量矩阵训练得到语义生成模型的方式具体为:
    获取所述历史故障描述文本对应的故障设备类型;
    根据所述词向量矩阵和所述类别标签训练分类模型,所述类别标签包括所述故障设备类型;
    根据所述分类模型得到语义生成模型。
  18. 根据权利要求17所述的装置,其特征在于,所述训练模块根据所述词向量矩阵和所述类别标签训练分类模型的方式具体为:
    将所述词向量矩阵和所述类别标签输入神经网络进行迭代训练,在每次迭代训练时对输入所述神经网络的词向量矩阵中的词向量和所述神经网络的参数进行调整,以生成所述分类模型。
  19. 一种计算机程序产品,其特征在于,当其在计算机上运行时,使得计算机执行上 述1~9中任意一项所述的方法。
  20. 一种计算机可读存储介质,其特征在于,计算机可读存储介质中存储有指令,当其在计算机上运行时,使得计算机执行上述1~9中任意一项所述的方法。
PCT/CN2019/084814 2018-05-03 2019-04-28 一种信息输出方法及装置 WO2019210820A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810415523.0A CN109902283B (zh) 2018-05-03 2018-05-03 一种信息输出方法及装置
CN201810415523.0 2018-05-03

Publications (1)

Publication Number Publication Date
WO2019210820A1 true WO2019210820A1 (zh) 2019-11-07

Family

ID=66943185

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/084814 WO2019210820A1 (zh) 2018-05-03 2019-04-28 一种信息输出方法及装置

Country Status (2)

Country Link
CN (1) CN109902283B (zh)
WO (1) WO2019210820A1 (zh)

Cited By (30)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110909550A (zh) * 2019-11-13 2020-03-24 北京环境特性研究所 文本处理方法、装置、电子设备和可读存储介质
CN111078822A (zh) * 2019-11-29 2020-04-28 北京百卓网络技术有限公司 基于中文小说文本的阅读器信息抽取方法及系统
CN111124925A (zh) * 2019-12-25 2020-05-08 斑马网络技术有限公司 基于大数据的场景提取方法、装置、设备和存储介质
CN111274366A (zh) * 2020-03-25 2020-06-12 联想(北京)有限公司 搜索推荐方法及装置、设备、存储介质
CN111291564A (zh) * 2020-03-03 2020-06-16 腾讯科技(深圳)有限公司 一种用于词向量获取的模型训练方法、装置和存储介质
CN111651601A (zh) * 2020-06-02 2020-09-11 全球能源互联网研究院有限公司 用于电力信息系统的故障分类模型的训练方法及分类方法
CN111767721A (zh) * 2020-03-26 2020-10-13 北京沃东天骏信息技术有限公司 信息处理方法、装置及设备
CN112069833A (zh) * 2020-09-01 2020-12-11 北京声智科技有限公司 日志分析方法、日志分析装置及电子设备
CN112463378A (zh) * 2020-11-27 2021-03-09 北京浪潮数据技术有限公司 一种服务器资产扫描方法、系统、电子设备及存储介质
CN112507720A (zh) * 2020-11-12 2021-03-16 西安交通大学 基于因果语义关系传递的图卷积网络根因识别方法
CN112529104A (zh) * 2020-12-23 2021-03-19 东软睿驰汽车技术(沈阳)有限公司 一种车辆故障预测模型生成方法、故障预测方法及装置
CN112711947A (zh) * 2021-01-09 2021-04-27 国网湖北省电力有限公司电力科学研究院 一种基于文本向量化的故障停电抢修事中处置参考方法
CN112749553A (zh) * 2020-06-05 2021-05-04 腾讯科技(深圳)有限公司 视频文件的文本信息处理方法、装置和服务器
CN112818008A (zh) * 2021-01-21 2021-05-18 中广核工程有限公司 核电调试故障智能诊断的方法、系统、介质及电子设备
CN112988921A (zh) * 2019-12-13 2021-06-18 北京四维图新科技股份有限公司 地图信息变化的识别方法和装置
CN113495949A (zh) * 2020-03-18 2021-10-12 北京沃东天骏信息技术有限公司 文本识别的方法、系统、计算机系统及介质
CN113591477A (zh) * 2021-08-10 2021-11-02 平安银行股份有限公司 基于关联数据的故障定位方法、装置、设备及存储介质
CN113610112A (zh) * 2021-07-09 2021-11-05 中国商用飞机有限责任公司上海飞机设计研究院 飞机装配质量缺陷辅助决策方法
CN113779975A (zh) * 2020-06-10 2021-12-10 北京猎户星空科技有限公司 一种语义识别方法、装置、设备及介质
CN113822016A (zh) * 2020-06-19 2021-12-21 阿里巴巴集团控股有限公司 文本数据处理方法、装置、电子设备及可读存储介质
CN113821418A (zh) * 2021-06-24 2021-12-21 腾讯科技(深圳)有限公司 故障跟因分析方法及装置、存储介质和电子设备
CN113961708A (zh) * 2021-11-10 2022-01-21 北京邮电大学 一种基于多层次图卷积网络的电力设备故障溯源方法
CN114036293A (zh) * 2021-11-03 2022-02-11 腾讯科技(深圳)有限公司 数据处理方法、装置及电子设备
CN114218402A (zh) * 2021-12-17 2022-03-22 迈创企业管理服务股份有限公司 一种计算机硬件故障更换件推荐的方法
CN114265930A (zh) * 2021-11-19 2022-04-01 国电南京自动化股份有限公司 一种基于事件抽取的低压用户故障报修合并处理方法
CN115687031A (zh) * 2022-11-15 2023-02-03 北京优特捷信息技术有限公司 一种告警描述文本的生成方法、装置、设备及介质
CN115994217A (zh) * 2022-11-29 2023-04-21 南京审计大学 一种财务报告舞弊检测方法及系统
CN116502058A (zh) * 2023-06-28 2023-07-28 长园深瑞能源技术有限公司 应用于充电桩系统的ai故障检测分析方法、系统及云平台
CN116738323A (zh) * 2023-08-08 2023-09-12 北京全路通信信号研究设计院集团有限公司 铁路信号设备的故障诊断方法、装置、设备及介质
CN117493886A (zh) * 2023-11-16 2024-02-02 重庆邮电大学 基于文本的变压器故障智能识别模型的训练方法及装置

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110245233A (zh) * 2019-06-19 2019-09-17 北京航空航天大学 一种故障确定方法及装置
CN110378486B (zh) * 2019-07-15 2021-12-03 出门问问信息科技有限公司 网络嵌入方法、装置、电子设备和存储介质
CN110569330A (zh) * 2019-07-18 2019-12-13 华瑞新智科技(北京)有限公司 一种基于智能选词的文本标注系统及装置、设备、介质
CN112667805B (zh) * 2019-09-30 2024-04-09 北京沃东天骏信息技术有限公司 一种工单类别确定方法、装置、设备及介质
CN111429155A (zh) * 2020-03-25 2020-07-17 中国银行股份有限公司 银行卡争议处理方法及装置
CN111858725A (zh) * 2020-04-30 2020-10-30 北京嘀嘀无限科技发展有限公司 一种事件属性确定方法及系统
CN112183994B (zh) * 2020-09-23 2023-05-12 南方电网数字电网研究院有限公司 一种设备状态的评估方法、装置、计算机设备和存储介质
CN112383421B (zh) * 2020-11-03 2023-03-24 中国联合网络通信集团有限公司 一种故障定位方法及装置
CN112925668B (zh) * 2021-02-25 2024-04-05 北京百度网讯科技有限公司 服务器健康评价方法、装置、设备以及存储介质
CN113657022B (zh) * 2021-07-15 2024-05-14 华为技术有限公司 一种芯片故障识别方法及相关设备
CN113722494A (zh) * 2021-09-10 2021-11-30 中国航空工业集团公司西安飞行自动控制研究所 一种基于自然语言理解的设备故障定位方法
CN114996391A (zh) * 2022-04-12 2022-09-02 支付宝(杭州)信息技术有限公司 故障变更定位方法、装置、设备、介质及程序产品

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103617157A (zh) * 2013-12-10 2014-03-05 东北师范大学 基于语义的文本相似度计算方法
CN103744905A (zh) * 2013-12-25 2014-04-23 新浪网技术(中国)有限公司 垃圾邮件判定方法和装置
CN104361026A (zh) * 2014-10-22 2015-02-18 北京航空航天大学 一种fmea分析过程中的故障知识存储和推送方法
CN107291699A (zh) * 2017-07-04 2017-10-24 湖南星汉数智科技有限公司 一种句子语义相似度计算方法
CN107704563A (zh) * 2017-09-29 2018-02-16 广州多益网络股份有限公司 一种问句推荐方法及系统

Family Cites Families (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2003173286A (ja) * 2001-12-05 2003-06-20 Nippon Telegr & Teleph Corp <Ntt> 分散ネットワークにおける意味情報取得方法
CN101795210A (zh) * 2010-01-11 2010-08-04 浪潮通信信息系统有限公司 一种处理通信网络故障的方法
KR101078751B1 (ko) * 2011-02-23 2011-11-02 한국과학기술정보연구원 어휘자원 시맨틱 네트워크의 관계 오류 검출 방법 및 장치
US20120233112A1 (en) * 2011-03-10 2012-09-13 GM Global Technology Operations LLC Developing fault model from unstructured text documents
US9256595B2 (en) * 2011-10-28 2016-02-09 Sap Se Calculating term similarity using a meta-model semantic network
CN102650960B (zh) * 2012-03-31 2015-04-15 北京奇虎科技有限公司 一种消除终端设备故障的方法及装置
US20130339787A1 (en) * 2012-06-15 2013-12-19 International Business Machines Coporation Systematic failure remediation
CN106815252B (zh) * 2015-12-01 2020-08-25 阿里巴巴集团控股有限公司 一种搜索方法和设备
CN107171819B (zh) * 2016-03-07 2020-02-14 北京华为数字技术有限公司 一种网络故障诊断方法及装置
CN106326346A (zh) * 2016-08-06 2017-01-11 上海高欣计算机系统有限公司 文本分类方法及终端设备
CN106941423B (zh) * 2017-04-13 2018-06-05 腾讯科技(深圳)有限公司 故障原因定位方法及装置
CN107248927B (zh) * 2017-05-02 2020-06-09 华为技术有限公司 故障定位模型的生成方法、故障定位方法和装置
CN107291693B (zh) * 2017-06-15 2021-01-12 广州赫炎大数据科技有限公司 一种改进词向量模型的语义计算方法
CN107340766B (zh) * 2017-07-10 2019-04-12 浙江大学 基于相似度的电力调度告警信号文本归类及故障诊断方法
CN107391727B (zh) * 2017-08-01 2020-03-06 北京航空航天大学 设备故障序列模式的挖掘方法及装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103617157A (zh) * 2013-12-10 2014-03-05 东北师范大学 基于语义的文本相似度计算方法
CN103744905A (zh) * 2013-12-25 2014-04-23 新浪网技术(中国)有限公司 垃圾邮件判定方法和装置
CN104361026A (zh) * 2014-10-22 2015-02-18 北京航空航天大学 一种fmea分析过程中的故障知识存储和推送方法
CN107291699A (zh) * 2017-07-04 2017-10-24 湖南星汉数智科技有限公司 一种句子语义相似度计算方法
CN107704563A (zh) * 2017-09-29 2018-02-16 广州多益网络股份有限公司 一种问句推荐方法及系统

Cited By (49)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110909550A (zh) * 2019-11-13 2020-03-24 北京环境特性研究所 文本处理方法、装置、电子设备和可读存储介质
CN110909550B (zh) * 2019-11-13 2023-11-03 北京环境特性研究所 文本处理方法、装置、电子设备和可读存储介质
CN111078822A (zh) * 2019-11-29 2020-04-28 北京百卓网络技术有限公司 基于中文小说文本的阅读器信息抽取方法及系统
CN112988921A (zh) * 2019-12-13 2021-06-18 北京四维图新科技股份有限公司 地图信息变化的识别方法和装置
CN111124925A (zh) * 2019-12-25 2020-05-08 斑马网络技术有限公司 基于大数据的场景提取方法、装置、设备和存储介质
CN111124925B (zh) * 2019-12-25 2024-04-05 斑马网络技术有限公司 基于大数据的场景提取方法、装置、设备和存储介质
CN111291564B (zh) * 2020-03-03 2023-10-31 腾讯科技(深圳)有限公司 一种用于词向量获取的模型训练方法、装置和存储介质
CN111291564A (zh) * 2020-03-03 2020-06-16 腾讯科技(深圳)有限公司 一种用于词向量获取的模型训练方法、装置和存储介质
CN113495949A (zh) * 2020-03-18 2021-10-12 北京沃东天骏信息技术有限公司 文本识别的方法、系统、计算机系统及介质
CN111274366A (zh) * 2020-03-25 2020-06-12 联想(北京)有限公司 搜索推荐方法及装置、设备、存储介质
CN111767721A (zh) * 2020-03-26 2020-10-13 北京沃东天骏信息技术有限公司 信息处理方法、装置及设备
CN111651601B (zh) * 2020-06-02 2023-04-18 全球能源互联网研究院有限公司 用于电力信息系统的故障分类模型的训练方法及分类方法
CN111651601A (zh) * 2020-06-02 2020-09-11 全球能源互联网研究院有限公司 用于电力信息系统的故障分类模型的训练方法及分类方法
CN112749553A (zh) * 2020-06-05 2021-05-04 腾讯科技(深圳)有限公司 视频文件的文本信息处理方法、装置和服务器
CN112749553B (zh) * 2020-06-05 2023-07-25 腾讯科技(深圳)有限公司 视频文件的文本信息处理方法、装置和服务器
CN113779975A (zh) * 2020-06-10 2021-12-10 北京猎户星空科技有限公司 一种语义识别方法、装置、设备及介质
CN113779975B (zh) * 2020-06-10 2024-03-01 北京猎户星空科技有限公司 一种语义识别方法、装置、设备及介质
CN113822016A (zh) * 2020-06-19 2021-12-21 阿里巴巴集团控股有限公司 文本数据处理方法、装置、电子设备及可读存储介质
CN113822016B (zh) * 2020-06-19 2024-03-22 阿里巴巴集团控股有限公司 文本数据处理方法、装置、电子设备及可读存储介质
CN112069833A (zh) * 2020-09-01 2020-12-11 北京声智科技有限公司 日志分析方法、日志分析装置及电子设备
CN112069833B (zh) * 2020-09-01 2024-04-30 北京声智科技有限公司 日志分析方法、日志分析装置及电子设备
CN112507720A (zh) * 2020-11-12 2021-03-16 西安交通大学 基于因果语义关系传递的图卷积网络根因识别方法
CN112463378A (zh) * 2020-11-27 2021-03-09 北京浪潮数据技术有限公司 一种服务器资产扫描方法、系统、电子设备及存储介质
CN112463378B (zh) * 2020-11-27 2023-12-22 北京浪潮数据技术有限公司 一种服务器资产扫描方法、系统、电子设备及存储介质
CN112529104A (zh) * 2020-12-23 2021-03-19 东软睿驰汽车技术(沈阳)有限公司 一种车辆故障预测模型生成方法、故障预测方法及装置
CN112711947A (zh) * 2021-01-09 2021-04-27 国网湖北省电力有限公司电力科学研究院 一种基于文本向量化的故障停电抢修事中处置参考方法
CN112711947B (zh) * 2021-01-09 2023-08-22 国网湖北省电力有限公司电力科学研究院 一种基于文本向量化的故障停电抢修事中处置参考方法
CN112818008A (zh) * 2021-01-21 2021-05-18 中广核工程有限公司 核电调试故障智能诊断的方法、系统、介质及电子设备
CN113821418B (zh) * 2021-06-24 2024-05-14 腾讯科技(深圳)有限公司 故障根因分析方法及装置、存储介质和电子设备
CN113821418A (zh) * 2021-06-24 2021-12-21 腾讯科技(深圳)有限公司 故障跟因分析方法及装置、存储介质和电子设备
CN113610112B (zh) * 2021-07-09 2024-04-16 中国商用飞机有限责任公司上海飞机设计研究院 飞机装配质量缺陷辅助决策方法
CN113610112A (zh) * 2021-07-09 2021-11-05 中国商用飞机有限责任公司上海飞机设计研究院 飞机装配质量缺陷辅助决策方法
CN113591477A (zh) * 2021-08-10 2021-11-02 平安银行股份有限公司 基于关联数据的故障定位方法、装置、设备及存储介质
CN113591477B (zh) * 2021-08-10 2023-09-15 平安银行股份有限公司 基于关联数据的故障定位方法、装置、设备及存储介质
CN114036293A (zh) * 2021-11-03 2022-02-11 腾讯科技(深圳)有限公司 数据处理方法、装置及电子设备
CN114036293B (zh) * 2021-11-03 2023-06-06 腾讯科技(深圳)有限公司 数据处理方法、装置及电子设备
CN113961708A (zh) * 2021-11-10 2022-01-21 北京邮电大学 一种基于多层次图卷积网络的电力设备故障溯源方法
CN113961708B (zh) * 2021-11-10 2024-04-23 北京邮电大学 一种基于多层次图卷积网络的电力设备故障溯源方法
CN114265930A (zh) * 2021-11-19 2022-04-01 国电南京自动化股份有限公司 一种基于事件抽取的低压用户故障报修合并处理方法
CN114218402A (zh) * 2021-12-17 2022-03-22 迈创企业管理服务股份有限公司 一种计算机硬件故障更换件推荐的方法
CN114218402B (zh) * 2021-12-17 2024-05-28 迈创企业管理服务股份有限公司 一种计算机硬件故障更换件推荐的方法
CN115687031A (zh) * 2022-11-15 2023-02-03 北京优特捷信息技术有限公司 一种告警描述文本的生成方法、装置、设备及介质
CN115994217B (zh) * 2022-11-29 2024-01-23 南京审计大学 一种财务报告舞弊检测方法及系统
CN115994217A (zh) * 2022-11-29 2023-04-21 南京审计大学 一种财务报告舞弊检测方法及系统
CN116502058A (zh) * 2023-06-28 2023-07-28 长园深瑞能源技术有限公司 应用于充电桩系统的ai故障检测分析方法、系统及云平台
CN116502058B (zh) * 2023-06-28 2023-09-26 长园深瑞能源技术有限公司 应用于充电桩系统的ai故障检测分析方法、系统及云平台
CN116738323B (zh) * 2023-08-08 2023-10-27 北京全路通信信号研究设计院集团有限公司 铁路信号设备的故障诊断方法、装置、设备及介质
CN116738323A (zh) * 2023-08-08 2023-09-12 北京全路通信信号研究设计院集团有限公司 铁路信号设备的故障诊断方法、装置、设备及介质
CN117493886A (zh) * 2023-11-16 2024-02-02 重庆邮电大学 基于文本的变压器故障智能识别模型的训练方法及装置

Also Published As

Publication number Publication date
CN109902283B (zh) 2023-06-06
CN109902283A (zh) 2019-06-18

Similar Documents

Publication Publication Date Title
WO2019210820A1 (zh) 一种信息输出方法及装置
US20210099336A1 (en) Fault root cause analysis method and apparatus
CN112633010B (zh) 基于多头注意力和图卷积网络的方面级情感分析方法及系统
CN113094200B (zh) 一种应用程序的故障预测方法和装置
Redl et al. Automatic SLA matching and provider selection in grid and cloud computing markets
CN112182219A (zh) 一种基于日志语义分析的在线服务异常检测方法
CN114785666B (zh) 一种网络故障排查方法与系统
TWI590095B (zh) 軟體功能驗證系統及其驗證方法
EP3846034B1 (en) Systems and methods for automated testing using artificial intelligence techniques
JP2023534696A (ja) ネットワークトポロジーにおけるアノマリー検知
CN111104242A (zh) 基于深度学习的操作系统的异常日志的处理方法及装置
US20210081612A1 (en) Relationship estimation model learning device, method, and program
WO2008005637A2 (en) Hybrid assessment tool, and systems and methods of quantifying risk
CN112015890B (zh) 电影剧本摘要的生成方法和装置
JP2018124829A (ja) 状態判定装置、状態判定方法及びプログラム
CN113824575B (zh) 故障节点识别的方法、装置、计算设备及计算机存储介质
CN112069833B (zh) 日志分析方法、日志分析装置及电子设备
KR101863569B1 (ko) 머신 러닝 기반의 취약점 정보를 분류하는 방법 및 장치
CN115658546A (zh) 一种基于异质信息网络的软件故障预测方法及系统
CN111562943B (zh) 一种基于事件嵌入树及gat网络的代码克隆检测方法和装置
KR101893029B1 (ko) 머신 러닝 기반의 취약점 정보를 분류하는 방법 및 장치
Escovedo et al. Learning under concept drift using a neuro-evolutionary ensemble
US20130325365A1 (en) Buildable part pairs in an unconfigured product structure
CN117254980B (zh) 一种基于注意力机制的工业网络安全风险评估方法及系统
CN115249133B (zh) 一种基于宽度学习网络的建筑施工过程风险分类方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19796720

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19796720

Country of ref document: EP

Kind code of ref document: A1