WO2021217843A1 - 企业舆情分析方法、装置、电子设备及介质 - Google Patents

企业舆情分析方法、装置、电子设备及介质 Download PDF

Info

Publication number
WO2021217843A1
WO2021217843A1 PCT/CN2020/098236 CN2020098236W WO2021217843A1 WO 2021217843 A1 WO2021217843 A1 WO 2021217843A1 CN 2020098236 W CN2020098236 W CN 2020098236W WO 2021217843 A1 WO2021217843 A1 WO 2021217843A1
Authority
WO
WIPO (PCT)
Prior art keywords
text
target
public opinion
warning
model
Prior art date
Application number
PCT/CN2020/098236
Other languages
English (en)
French (fr)
Inventor
周欣欣
李文
吴晓贝
于修铭
汪伟
Original Assignee
平安科技(深圳)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 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2021217843A1 publication Critical patent/WO2021217843A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/20Natural language analysis
    • G06F40/205Parsing
    • G06F40/216Parsing using statistical methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90Details of database functions independent of the retrieved data types
    • G06F16/95Retrieval from the web
    • G06F16/951Indexing; Web crawling techniques
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90Details of database functions independent of the retrieved data types
    • G06F16/95Retrieval from the web
    • G06F16/953Querying, e.g. by the use of web search engines
    • G06F16/9535Search customisation based on user profiles and personalisation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/20Natural language analysis
    • G06F40/279Recognition of textual entities
    • G06F40/289Phrasal analysis, e.g. finite state techniques or chunking
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/30Semantic analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/049Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/20Image preprocessing
    • G06V10/22Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to an enterprise public opinion analysis method, device, electronic equipment and medium.
  • the inventor realizes that the existing model or application system has a relatively single analysis of public opinion information, which leads to a large amount of public opinion information that is not fully utilized, and thus it is impossible to send early warning information to enterprises in a timely and accurate manner.
  • the first aspect of this application provides an enterprise public opinion analysis method, and the enterprise public opinion analysis method includes:
  • a second aspect of the present application provides an electronic device including a processor and a memory, and the processor is configured to execute computer-readable instructions stored in the memory to implement the following steps:
  • a third aspect of the present application provides a computer-readable storage medium having at least one computer-readable instruction stored thereon, and the at least one computer-readable instruction is executed by a processor to implement the following steps:
  • the fourth aspect of the present application provides an enterprise public opinion analysis device, and the enterprise public opinion analysis device includes:
  • the obtaining unit is used to obtain public opinion text from a preset data channel
  • the preprocessing unit is used to preprocess the obtained public opinion text to obtain multiple phrases of each public opinion text;
  • the determining unit is used to perform named entity recognition on multiple phrases of each public opinion text, and determine the public opinion text recognized by the corporate entity as the target text;
  • the classification unit is configured to perform subject recognition on the target text based on the dependency syntax analysis technology to obtain a recognition result, and classify the target text according to a preset classification standard to obtain a classification result;
  • An input unit configured to input the target text, the recognition result, and the classification result into a pre-trained emotion model to obtain an emotion result
  • the selection unit is used to select the target text whose emotional result is a negative emotion as the warning text;
  • the determining unit is further configured to match the warning text with the warning phrase in the configuration list, and determine the matched warning phrase as the warning category of the warning text;
  • the generating unit is used to obtain the target company of the warning text, and generate the warning information of the target company according to the warning category and the warning text.
  • this application uses artificial intelligence, can obtain public opinion texts from massive data for natural language processing, and send early warning information to enterprises in a timely and accurate manner by making full use of public opinion information.
  • this application is also applied to the smart city fields such as smart medical care, and the response policy is determined by analyzing public opinion information such as the epidemic, so as to promote the development of smart cities.
  • Fig. 1 is a flowchart of a preferred embodiment of a method for analyzing public opinion of an enterprise disclosed in the present application.
  • Fig. 2 is a functional module diagram of a preferred embodiment of an enterprise public opinion analysis device disclosed in the present application.
  • FIG. 3 is a schematic diagram of the structure of an electronic device according to a preferred embodiment of the method for analyzing enterprise public opinion according to the present application.
  • the enterprise public opinion analysis method of the embodiment of the present application is applied to an electronic device, and can also be applied to a hardware environment composed of an electronic device and a server connected to the electronic device through a network, and is executed by the server and the electronic device.
  • Networks include, but are not limited to: wide area networks, metropolitan area networks, or local area networks.
  • FIG. 1 is a flowchart of a preferred embodiment of a method for analyzing public opinion of an enterprise disclosed in the present application. Among them, according to different needs, the order of the steps in the flowchart can be changed, and some steps can be omitted.
  • S10 Obtain public opinion text from a preset data channel.
  • the preset data channels include, but are not limited to: newspapers, Weibo, WeChat, forums, etc.
  • the public opinion texts obtained from the aforementioned data channels mainly include news, forum posts, and microblogs. Blog posts, WeChat articles, etc.
  • the electronic device obtaining public opinion text from a preset data channel includes:
  • the electronic device uses optical character recognition (Optical Character Recognition, OCR) to scan and recognize paper text, and use the scanned electronic text as the public opinion text.
  • OCR Optical Character Recognition
  • the electronic device acquires the text published by the user in the social software from the open interface of the social software based on the web crawler technology as the public opinion text.
  • the electronic device obtains the text published by the user in the portal website through the anonymous proxy pool as the public opinion text.
  • the electronic device obtaining the text published by the user in the portal website through the anonymous proxy pool includes:
  • the anonymity proxy pool is formed based on a normal HTTP (HyperText Transfer Protocol) proxy and a high anonymity proxy, and the electronic device randomly selects a variable address from the addresses generated by the normal HTTP proxy and the high anonymity proxy Further, the electronic device splices the preset header file with the variable address, and uses the spliced address as a crawler to obtain the text published by the user from the portal website.
  • HTTP HyperText Transfer Protocol
  • the probability of the crawler being recognized by the portal website can be reduced.
  • the electronic device preprocesses the obtained public opinion text to obtain multiple phrases corresponding to each public opinion text including:
  • the electronic device filters the special characters and stop words in the public opinion text to obtain the first text, and the electronic device performs deduplication processing on the first text based on the cosine distance formula to obtain the second text.
  • the electronic device segments the second text according to a preset custom dictionary to obtain a segmentation position, and constructs at least one directed acyclic graph according to the segmentation position.
  • the electronic device segments the second text according to the custom
  • the weights in the dictionary calculate the probability of each directed acyclic graph, and determine the segmentation position corresponding to the directed acyclic graph with the highest probability as the target segmentation position, and the electronic device determines the target segmentation position according to the target segmentation position. Multiple phrases for each public opinion text.
  • the special characters include, but are not limited to: emoticons, symbol patterns, etc.
  • At least one custom word and a weight corresponding to each custom word are stored in the preset custom dictionary.
  • the repeated text in the first text can be removed, and the time spent in processing the repeated text can be avoided; by determining the segmentation position with the highest probability as the target segmentation position, the public opinion text can be accurately segmented.
  • the electronic device performs de-duplication processing on the first text based on the cosine distance formula to obtain the second text including:
  • the electronic device calculates the hash value of the first text according to the title of the first text, the electronic device extracts preset features from the first text and establishes a feature index, and according to the first text
  • the hash value of the text is calculated by using the cosine distance formula to calculate the similarity distance of any two first texts in the first text to obtain the similarity distance of the text pair, where the text pair includes any two first texts, and
  • the electronic device searches for a text pair whose similarity distance is greater than a preset value through the feature index, and determines the text pair as a similar text pair.
  • the electronic device determines whether the preset features in the similar text pair are the same. When the preset features in the similar text pairs are the same, the electronic device removes any first text in the similar text pairs, and determines the retained first text as the second text.
  • S12 Perform named entity recognition on multiple phrases of each public opinion text, and determine the public opinion text recognized as the target text of the corporate entity.
  • that the electronic device performs named entity recognition on multiple phrases of each public opinion text, and determining the public opinion text identified by the corporate entity as the target text includes:
  • the electronic device performs vectorization processing on multiple phrases of each public opinion text to obtain the input vector sequence of each public opinion text. Further, the electronic device inputs the input vector sequence of each public opinion text into the NER model, And obtain the output probability and transition probability of each tag corresponding to each sequence position in the activation layer. For each sequence position, the electronic device performs a weighted sum operation on the output probability and transition probability of each tag to obtain each The value of the label, the electronic device determines the label with the highest value as the output label at each sequence position, and combines the output labels at each sequence position to obtain the entity list of each public opinion text.
  • the electronic device determines the public opinion text corresponding to the entity list as the target text, or when it detects that the entity list contains multiple corporate entities, the electronic device determines the public opinion corresponding to the entity list
  • the text is determined to be a multi-text, and the multi-text is split according to the multiple enterprise entities to obtain the target text.
  • each tag e.g., tag B-PER
  • B-PER represents the first character of the person's name Mark
  • E-PER means the end character mark of the name
  • O means the independent character mark
  • B-COM means the first character mark of the business name
  • I-COM means the middle character mark of the business name
  • E-COM means the end character mark of the business name
  • the representation form of the entity list A is (B-PER, E-PER, O, O, B-COM, I-COM, E-COM)
  • the electronic device determines that the entity list A contains an enterprise entity
  • the corporate entity is (B-COM, I-COM, E-COM)
  • the representation of entity list B is (B-COM1, E-COM1, O, O, B-COM2, I-COM2, E-COM2)
  • the electronic device determines that the entity list B contains two business entities, and the two business entities are (B-COM1, E-COM1) and (B-COM2, I-COM2, E-COM2).
  • the enterprise entity corresponding to (B-COM1, E-COM1) can be "T company”; the enterprise entity corresponding to (B-COM2, I-COM2, E-COM2) can be "G company”.
  • the public opinion text with the corporate entity can be accurately determined, avoiding the failure to determine the corporate entity during subsequent subject identification, and then the redundant public opinion text can be processed.
  • the method before inputting the input vector sequence of each public opinion text into the NER model, the method further includes:
  • the electronic device obtains corporate public opinion data, and selects a target labeling mode according to the corporate public opinion data, the electronic device adds the target labeling mode to the combined model to obtain a labeling model, and the electronic device responds to the labeling model Cutting is performed to obtain a cutting model, and the electronic device reduces the order of the cutting model to obtain the NER model.
  • the combined model includes:
  • the electronic device cutting the annotation model to obtain the cutting model includes:
  • the electronic device extracts all convolution kernels from the annotation model, and further, the electronic device uses the gray correlation analysis method to quantify the importance of each convolution kernel in all the convolution kernels to obtain each
  • the quantized value of the importance of the convolution kernel, and all the convolution kernels are sorted according to the magnitude of the quantized value in ascending order to obtain a queue, and the electronic device selects the top N from the queue
  • a convolution kernel is used as a target convolution kernel, the N is a positive integer, and the electronic device deletes the target convolution kernel in the annotation model to obtain the crop model.
  • the model can be tailored while ensuring the accuracy of the model, and furthermore, the efficiency of named entity recognition can be improved.
  • the electronic device performs vectorization processing on multiple phrases of each public opinion text, and obtaining the input vector sequence of each public opinion text includes:
  • the electronic device obtains the coding vector of each phrase of each public opinion text according to the preset coding table, and generates the position vector of each phrase according to the position number of each phrase of each public opinion text, and the electronic device splices each The encoding vector of the phrase and the position vector of each phrase are used to obtain the target vector of each phrase.
  • the electronic device combines the target vector of each phrase of each public opinion text according to the word order to obtain the input vector sequence of each public opinion text.
  • the target vector of each phrase is generated, so that the generated target vector has contextual semantic characteristics, thereby improving the accuracy of phrase named entity recognition.
  • this embodiment may also store the public opinion text in a node of a blockchain.
  • the public opinion text corresponding to the entity list is determined to be a multi-text, and the multi-text is split according to the multiple corporate entities, Obtain the target text.
  • Public opinion text 1 is: A company is a well-known company. The sales volume of B company is leading in the industry.
  • the electronic device determines the public opinion text 1 as a multi-text, and splits the public opinion text 1 according to the A corporate entity and the B corporate entity to obtain the target text 2 is: A company is a well-known company; target text 3 is: B company’s sales volume is leading in the industry.
  • S13 Perform subject discrimination on the target text based on the dependency syntax analysis technology to obtain a discrimination result recognition result, and classify the target text according to a preset classification standard to obtain a classification result.
  • the identification result of the discrimination result includes: the company is the subject in the target text, or the company is mentioned in the target text.
  • the classification results include: recruitment, advertising, consulting, etc.
  • the electronic device may classify the target text from multiple dimensions, and each dimension has multiple classification results.
  • the electronic device performs subject recognition on the target text based on dependency syntax analysis technology, and obtaining the recognition result includes:
  • the electronic device obtains the core verb of each text sentence in the target text according to the dependency syntax analysis technology, the electronic device determines the participle of the main predicate relationship and the verb-object relationship in the dependency relationship with the core verb, and the electronic device The device calculates the total number of the word segmentation in the target text, the electronic device acquires the target word segmentation corresponding to the business entity, and calculates the target number of the target word segmentation in the target text, and the electronic device The target number is divided by the total number to obtain a target ratio.
  • the electronic device determines the target word segmentation as the main body of the target text, as the The recognition result, or when it is detected that the target ratio is less than a second preset ratio, the electronic device determines the target word segmentation as a mention of the target text as the recognition result.
  • the core verb is "make”, and then according to the dependency syntactic analysis technology, find the words whose dependency relationship with the core verb is "subject-predicate relationship” and "verb-object relationship”, which are "A enterprise” and “notify” respectively.
  • the proportion of the corresponding target word segmentation of the enterprise in the target text is greater than the first preset ratio.
  • the corresponding The proportion of the target word segmentation in the target text is smaller than the second preset proportion.
  • the core verbs and participles can be quickly determined through the dependency syntax analysis technology, and then the recognition results can be accurately determined by detecting the proportion of the target participles.
  • the emotional result includes: neutral emotion, negative emotion, and positive emotion.
  • the method before the target text, the recognition result of the discrimination result, and the classification result are input into a pre-trained emotion model, the method further includes:
  • the electronic device uses a crawler program to obtain the first historical data of all classification results, the electronic device inputs the first historical data to the forgetting gate layer for forgetting processing, and obtains training data, and the electronic device uses a cross-validation method to
  • the training data is divided into a training set and a verification set.
  • the electronic device inputs the data in the training set to the input gate layer for training to obtain a primary learner.
  • the electronic device adjusts all the data in the verification set.
  • the primary learner obtains a secondary learner
  • the electronic device obtains second historical data according to the classification result of the target text, and uses the second historical data as test data to test the secondary learner to obtain a test
  • the electronic device calculates the target number of the second historical data that passed the test, and calculates the total number of the second historical data participating in the test, and the electronic device divides the target number by the total number to obtain the test success
  • the electronic device determines the secondary learner as the emotional model, or when the test success rate is less than or equal to the configured value
  • the electronic device adjusts the secondary learner according to the second historical data to obtain the emotion model.
  • the trained emotion model can be made more accurate.
  • the electronic device when the emotional result of the public opinion text is a neutral emotion or a positive emotion, the electronic device does not need to warn the public opinion text.
  • a plurality of early warning phrases are stored in the configuration list, and the early warning phrases include: legal proceedings, personnel changes, police punishments, and the like.
  • S17 Acquire the target company of the warning text, and generate warning information of the target company according to the warning category and the warning text.
  • generating, by the electronic device, the early warning information of the target company according to the warning category and the warning text includes:
  • the electronic device extracts a strategy from a case database according to the warning category, and further, the electronic device generates the warning information according to the warning category, the warning text, and the strategy.
  • the above-mentioned early warning information may also be stored in a node of a blockchain.
  • public opinion text can be obtained from massive data for natural language processing, and early warning information can be sent to the enterprise in a timely and accurate manner by making full use of public opinion information.
  • this application applies artificial intelligence and smart healthcare to smart city fields, and determines response policies by analyzing public opinion information such as the epidemic, so as to promote the development of smart cities.
  • FIG. 2 is a functional module diagram of a preferred embodiment of an enterprise public opinion analysis device disclosed in the present application.
  • the enterprise public opinion analysis device runs in an electronic device.
  • the enterprise public opinion analysis device may include multiple functional modules composed of program code segments, and the program is a series of computer-readable instruction codes.
  • the program code of each program segment in the enterprise public opinion analysis device can be stored in a memory and executed by at least one processor to execute part or all of the steps in the enterprise public opinion analysis method described in FIG. 1. For details, please refer to The related description in the method shown in FIG. 1 will not be repeated here.
  • the enterprise public opinion analysis device can be divided into multiple functional modules according to the functions it performs.
  • the functional modules may include: acquisition unit 110, preprocessing unit 111, determination unit 112, classification unit 113, input unit 114, selection unit 115, generation unit 116, addition unit 117, cropping unit 118, order reduction unit 119, division The unit 120, the adjustment unit 121, the test unit 122, and the calculation unit 123.
  • the module referred to in this application refers to a series of computer-readable instruction segments that can be executed by at least one processor and can complete fixed functions, and are stored in a memory.
  • the obtaining unit 110 obtains the public opinion text from a preset data channel.
  • the preset data channels include, but are not limited to: newspapers, Weibo, WeChat, forums, etc.
  • the public opinion texts obtained from the aforementioned data channels mainly include news, forum posts, and microblogs. Blog posts, WeChat articles, etc.
  • the obtaining unit 110 obtaining public opinion text from a preset data channel includes:
  • the acquisition unit 110 scans and recognizes the paper text using optical character recognition (OCR) technology, and uses the scanned electronic text as the public opinion text.
  • OCR optical character recognition
  • the obtaining unit 110 obtains the text published by the user in the social software from the open interface of the social software based on the web crawler technology, as the public opinion text.
  • the obtaining unit 110 obtains the text published by the user in the portal website through the anonymous proxy pool as the public opinion text.
  • the obtaining unit 110 obtaining the text published by the user in the portal website through the anonymous proxy pool includes:
  • the anonymity proxy pool is formed based on a normal HTTP (HyperText Transfer Protocol) proxy and a high-anonymity proxy, and the acquisition unit 110 randomly selects and variable from addresses generated by the normal HTTP proxy and the high-anonymity proxy Address, further, the obtaining unit 110 splices a preset header file with the variable address, and uses the spliced address as a crawler to obtain the text published by the user from the portal website.
  • HTTP HyperText Transfer Protocol
  • the probability of the crawler being recognized by the portal website can be reduced.
  • the preprocessing unit 111 preprocesses the acquired public opinion text to obtain multiple phrases of each public opinion text.
  • the preprocessing unit 111 preprocesses the obtained public opinion text to obtain multiple phrases corresponding to each public opinion text including:
  • the preprocessing unit 111 filters the special characters and stop words in the public opinion text to obtain the first text, and the preprocessing unit 111 performs deduplication processing on the first text based on the cosine distance formula to obtain the second
  • the preprocessing unit 111 segments the second text according to a preset custom dictionary to obtain a segmentation position, and constructs at least one directed acyclic graph according to the segmentation position.
  • the processing unit 111 calculates the probability of each directed acyclic graph according to the weights in the custom dictionary, and determines the segmentation position corresponding to the directed acyclic graph with the highest probability as the target segmentation position.
  • the preprocessing The unit 111 determines multiple phrases of each public opinion text according to the target segmentation position.
  • the special characters include, but are not limited to: emoticons, symbol patterns, etc.
  • At least one custom word and a weight corresponding to each custom word are stored in the preset custom dictionary.
  • the repeated text in the first text can be removed, and the time spent in processing the repeated text can be avoided; by determining the segmentation position with the highest probability as the target segmentation position, the public opinion text can be accurately segmented.
  • the preprocessing unit 111 performs de-duplication processing on the first text based on the cosine distance formula to obtain the second text including:
  • the preprocessing unit 111 calculates the hash value of the first text according to the title of the first text, and the preprocessing unit 111 extracts preset features from the first text and establishes a feature index, and according to The hash value of the first text is calculated by using a cosine distance formula to calculate the similarity distance of any two first texts in the first text to obtain the similarity distance of the text pair, wherein the text pair includes any two first texts.
  • the preprocessing unit 111 searches for a text pair whose similarity distance is greater than a preset value through the feature index, and determines the text pair as a similar text pair.
  • the preprocessing unit 111 judges the text pair in the similar text pair Whether the preset features are the same, when the preset features in the similar text pair are the same, the preprocessing unit 111 removes any first text in the similar text pair, and determines the retained first text as the first text in the similar text pair. State the second text.
  • the determining unit 112 performs named entity recognition on multiple phrases of each public opinion text, and determines the public opinion text that has recognized the corporate entity as the target text.
  • the determining unit 112 performs named entity recognition on multiple phrases of each public opinion text, and determining the public opinion text recognized as the target text of the corporate entity includes:
  • the determining unit 112 performs vectorization processing on multiple phrases of each public opinion text to obtain the input vector sequence of each public opinion text. Further, the determining unit 112 inputs the input vector sequence of each public opinion text to the NER model , And obtain the output probability and transition probability of each tag corresponding to each sequence position in the activation layer. For each sequence position, the determining unit 112 performs a weighted sum operation on the output probability and transition probability of each tag, To obtain the value of each tag, the determination unit 112 determines the tag with the highest value as the output tag at each sequence position, and combines the output tags at each sequence position to obtain the entity list of each public opinion text.
  • the determining unit 112 determines the public opinion text corresponding to the entity list as the target text, or when multiple corporate entities are detected in the entity list, the determining unit 112 determines The public opinion text corresponding to the entity list is determined to be a multi-text, and the multi-text is split according to the multiple enterprise entities to obtain the target text.
  • the layer can obtain the output probability of each tag at the i-th sequence position, respectively: P iy , P iy ,..., P iym ; the determination unit 112 can obtain each of the i-th sequence positions from the activation layer transition probability labels, namely: a iy1, a iy2, ... , a iym.
  • each tag e.g., tag B-PER
  • B-PER represents the first character of the person's name Mark
  • E-PER means the end character mark of the name
  • O means the independent character mark
  • B-COM means the first character mark of the business name
  • I-COM means the middle character mark of the business name
  • E-COM means the end character mark of the business name
  • the representation form of the entity list A is (B-PER, E-PER, O, O, B-COM, I-COM, E-COM)
  • the electronic device determines that the entity list A contains an enterprise entity
  • the corporate entity is (B-COM, I-COM, E-COM)
  • the representation of entity list B is (B-COM1, E-COM1, O, O, B-COM2, I-COM2, E-COM2)
  • the determining unit 112 determines that the entity list B contains two business entities, and the two business entities are (B-COM1, E-COM1) and (B-COM2, I-COM2, E-COM2).
  • the enterprise entity corresponding to (B-COM1, E-COM1) can be "T company”; the enterprise entity corresponding to (B-COM2, I-COM2, E-COM2) can be "G company”.
  • the public opinion text with the corporate entity can be accurately determined, avoiding the failure to determine the corporate entity during subsequent subject identification, and then the redundant public opinion text can be processed.
  • the acquiring unit 110 acquires corporate public opinion data, and selects a target labeling mode according to the corporate public opinion data, and the adding unit 117 sets the target labeling mode It is added to the combined model to obtain the annotation model, the cropping unit 118 crops the annotation model to obtain the crop model, and the order reduction unit 119 reduces the order of the crop model to obtain the NER model.
  • the combined model includes:
  • the cropping unit 118 crops the annotation model, and obtaining the cropped model includes:
  • the cropping unit 118 extracts all convolution kernels from the annotation model, and further, the cropping unit 118 uses a gray correlation analysis method to quantify the importance of each convolution kernel in all the convolution kernels to obtain The quantized value of the importance of each convolution kernel, and all the convolution kernels are sorted in ascending order according to the size of the quantized value to obtain a queue, and the cropping unit 118 selects from the queue
  • the first N convolution kernels are used as target convolution kernels, where N is a positive integer, and the clipping unit 118 deletes the target convolution kernel in the annotation model to obtain the clipping model.
  • the model can be tailored while ensuring the accuracy of the model, and furthermore, the efficiency of named entity recognition can be improved.
  • the determining unit 112 performs vectorization processing on multiple phrases of each public opinion text, and obtaining the input vector sequence of each public opinion text includes:
  • the determining unit 112 obtains the coding vector of each phrase of each public opinion text according to the preset coding table, and generates the position vector of each phrase according to the position number of each phrase of each public opinion text, and the determining unit 112 concatenates The encoding vector of each phrase and the position vector of each phrase are used to obtain the target vector of each phrase.
  • the determining unit 112 combines the target vector of each phrase of each public opinion text according to the word order to obtain the input vector of each public opinion text sequence.
  • the target vector of each phrase is generated, so that the generated target vector has contextual semantic characteristics, thereby improving the accuracy of phrase named entity recognition.
  • the public opinion text corresponding to the entity list is determined to be a multi-text, and the multi-text is split according to the multiple corporate entities, Obtain the target text.
  • Public opinion text 1 is: A company is a well-known company. The sales volume of B company is leading in the industry.
  • the electronic device determines the public opinion text 1 as a multi-text, and splits the public opinion text 1 according to the A corporate entity and the B corporate entity to obtain the target text 2 is: A company is a well-known company; target text 3 is: B company’s sales volume is leading in the industry.
  • the classification unit 113 performs subject identification on the target text based on the dependency syntax analysis technology to obtain the identification result of the discrimination result, and classifies the target text according to a preset classification standard to obtain the classification result.
  • the identification result of the discrimination result includes: the company is the subject in the target text, or the company is mentioned in the target text.
  • the classification results include: recruitment, advertising, consulting, etc.
  • the classification unit 113 may classify the target text from multiple dimensions, and each dimension has multiple classification results.
  • the classification unit 113 performs subject recognition on the target text based on dependency syntax analysis technology, and obtaining the recognition result includes:
  • the classification unit 113 obtains the core verbs of each text sentence in the target text according to the dependency syntax analysis technology, and the classification unit 113 determines the participles of the main predicate relationship and the verb-object relationship with the core verb dependence relationship, so The classification unit 113 calculates the total number of the word segmentation in the target text, the classification unit 113 obtains the target word segmentation corresponding to the business entity, and calculates the target number of the target word segmentation in the target text. The classification unit 113 divides the target number by the total number to obtain a target ratio.
  • the classification unit 113 determines the target word segmentation as the target text As the recognition result, or when it is detected that the target proportion is less than a second preset proportion, the classification unit 113 determines the target word segmentation as a mention of the target text as the recognition result .
  • the core verb is "make”, and then according to the dependency syntactic analysis technology, find the words whose dependency relationship with the core verb is "subject-predicate relationship” and "verb-object relationship”, which are "A enterprise” and “notify” respectively.
  • the proportion of the corresponding target word segmentation of the enterprise in the target text is greater than the first preset ratio.
  • the corresponding The proportion of the target word segmentation in the target text is smaller than the second preset proportion.
  • the core verbs and participles can be quickly determined through the dependency syntax analysis technology, and then the recognition results can be accurately determined by detecting the proportion of the target participles.
  • the input unit 114 inputs the target text, the recognition result of the discrimination result, and the classification result into a pre-trained emotion model to obtain an emotion result.
  • the emotional result includes: neutral emotion, negative emotion, and positive emotion.
  • the acquisition unit 110 uses a crawler
  • the program obtains the first historical data of all the classification results
  • the input unit 114 inputs the first historical data to the forgetting gate layer for forgetting processing to obtain training data
  • the dividing unit 120 uses a cross-validation method to divide the training data into Training set and verification set.
  • the input unit 114 inputs the data in the training set to the input gate layer for training to obtain a primary learner.
  • the adjustment unit 121 adjusts the primary learner according to the data in the verification set to obtain The secondary learner, the acquiring unit 110 acquires second historical data according to the classification result of the target text, and the testing unit 122 uses the second historical data as the test data to test the secondary learner, obtains the test result, and calculates
  • the unit 123 calculates the target number of the second historical data that passed the test and calculates the total number of the second historical data participating in the test.
  • the calculation unit 123 divides the target number by the total number to obtain the test success rate.
  • the determining unit 112 determines the secondary learner as the emotional model, or when the test success rate is less than or equal to the configured value, the adjusting unit 121 The secondary learner is adjusted according to the second historical data to obtain the emotion model.
  • the trained emotion model can be made more accurate.
  • the selecting unit 115 selects the target text whose emotional result is a negative emotion as the warning text.
  • the determining unit 112 matches the warning text with the warning phrase in the configuration list, and determines the matched warning phrase as the warning category of the warning text.
  • a plurality of early warning phrases are stored in the configuration list, and the early warning phrases include: legal proceedings, personnel changes, police punishments, and the like.
  • the generating unit 116 obtains the target company of the warning text, and generates warning information of the target company according to the warning category and the warning text.
  • the generating unit 116 generating the early warning information of the target company according to the warning category and the warning text includes:
  • the generating unit 116 extracts a strategy from the case database according to the warning category. Further, the generating unit 116 generates the warning information according to the warning category, the warning text, and the strategy.
  • the above-mentioned early warning information may also be stored in a node of a blockchain.
  • public opinion text can be obtained from massive data for natural language processing, and early warning information can be sent to the enterprise in a timely and accurate manner by making full use of public opinion information.
  • this application applies artificial intelligence and smart healthcare to smart city fields, and determines response policies by analyzing public opinion information such as the epidemic, so as to promote the development of smart cities.
  • FIG. 3 is a schematic structural diagram of an electronic device according to a preferred embodiment of the method for analyzing enterprise public opinion according to the present application.
  • the electronic device 3 includes a memory 31, at least one processor 32, computer readable instructions 33 stored in the memory 31 and executable on the at least one processor 32, and at least one communication bus 34.
  • FIG. 3 is only an example of the electronic device 3, and does not constitute a limitation on the electronic device 3. It may include more or less components than those shown in the figure, or a combination. Certain components, or different components, for example, the electronic device 3 may also include input and output devices, network access devices, and so on.
  • the electronic device 3 also includes, but is not limited to, any electronic product that can interact with the user through a keyboard, a mouse, a remote control, a touch panel, or a voice control device, for example, a personal computer, a tablet computer, a smart phone, etc.
  • Personal digital assistants Personal Digital Assistant, PDA
  • game consoles interactive network television (Internet Protocol Television, IPTV), smart wearable devices, etc.
  • the at least one processor 32 may be a central processing unit (Central Processing Unit, CPU), or other general-purpose processors, digital signal processors (Digital Signal Processors, DSPs), and application specific integrated circuits (ASICs). ), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.
  • the processor 32 can be a microprocessor, or the processor 32 can also be any conventional processor, etc.
  • the processor 32 is the control center of the electronic device 3, and connects the entire electronic device 3 through various interfaces and lines. Parts.
  • the memory 31 may be used to store the computer-readable instructions 33 and/or modules/units, and the processor 32 can run or execute the computer-readable instructions and/or modules/units stored in the memory 31, and
  • the data stored in the memory 31 is called to realize various functions of the electronic device 3.
  • the memory 31 may mainly include a storage program area and a storage data area.
  • the storage program area may store an operating system, an application program required by at least one function (such as a sound playback function, an image playback function, etc.), etc.; the storage data area may Data (such as audio data) created according to the use of the electronic device 3 and the like are stored.
  • the memory 31 may include volatile memory such as high-speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), and a secure digital ( Secure Digital, SD card, Flash Card, at least one magnetic disk storage device, flash memory device, or other non-volatile solid state storage device.
  • volatile memory such as high-speed random access memory
  • non-volatile memory such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), and a secure digital ( Secure Digital, SD card, Flash Card, at least one magnetic disk storage device, flash memory device, or other non-volatile solid state storage device.
  • non-volatile memory such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), and a secure digital ( Secure Digital, SD card, Flash Card, at least one magnetic disk storage device, flash memory device, or other non-volatile solid state storage
  • the memory 31 in the electronic device 3 stores a plurality of instructions to implement a corporate public opinion analysis method
  • the processor 32 can execute the plurality of instructions so as to realize: obtain from a preset data channel Public opinion text; preprocess the obtained public opinion text to obtain multiple phrases of each public opinion text; perform named entity recognition on multiple phrases of each public opinion text, and determine the public opinion text recognized by the enterprise entity as the target text
  • the target text is subject to recognition to obtain the recognition result, and the target text is classified according to the preset classification criteria to obtain the classification result;
  • the target text, the recognition result and The classification result is input into the pre-trained emotional model to obtain the emotional result;
  • the target text whose emotional result is negative emotion is selected as the warning text;
  • the warning text is matched with the warning phrase in the configuration list, and the The matched warning phrase is determined as the warning category of the warning text; the target company of the warning text is obtained, and the warning information of the target company is generated according to the warning category and the warning text.
  • public opinion text can be obtained from massive data for natural language processing, and early warning information can be sent to the enterprise in a timely and accurate manner by making full use of public opinion information.
  • this application applies artificial intelligence and smart healthcare to smart city fields, and determines response policies by analyzing public opinion information such as the epidemic, so as to promote the development of smart cities.
  • the integrated module/unit of the electronic device 3 may be stored in a computer-readable storage medium, which may be non-easy.
  • a volatile storage medium can also be a volatile storage medium.
  • the computer-readable instruction includes computer-readable instruction code
  • the computer-readable instruction code may be in the form of source code, object code, executable file, or some intermediate form.
  • the computer-readable medium may include: any entity or device capable of carrying the computer-readable instruction code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory).
  • modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the objectives of the solutions of the embodiments.
  • the functional modules in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
  • the above-mentioned integrated unit may be implemented in the form of hardware, or may be implemented in the form of hardware plus software functional modules.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种企业舆情分析方法、装置、电子设备以及存储介质适用于金融科技领域。该方法获取舆情文本并进行预处理,得到多个词组,将识别到企业实体的舆情文本确定为目标文本,对目标文本进行主体识别,得到识别结果,对目标文本进行分类,得到分类结果,将目标文本、识别结果及分类结果输入至预先训练好的情感模型中,得到情感结果,选取出情感结果为负面情绪的目标文本,作为预警文本,将预警文本与配置列表中的预警词组进行匹配,将匹配到的预警词组确定为预警文本的预警类别,获取预警文本的目标企业并根据预警类别及预警文本生成预警信息。该方法能够及时、准确地发出预警。

Description

企业舆情分析方法、装置、电子设备及介质
本申请要求于2020年04月29日提交中国专利局,申请号为202010354452.5发明名称为“企业舆情分析方法、装置、电子设备及介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及人工智能技术领域,尤其涉及一种企业舆情分析方法、装置、电子设备及介质。
背景技术
随着微博、论坛等平台的出现,催生了大量的舆情信息,这些舆情信息中存在对企业有积极影响的信息,也存在对企业有负面影响的信息。通过对舆情信息的分析,可以及时找出对企业产生负面影响的舆情信息,方便企业决策者迅速制定应对策略,从而提高舆情信息的利用率和决策的正确性。
发明人意识到现有的模型或应用系统对舆情信息的分析比较单一,导致大量的舆情信息没有充分利用,进而无法及时、准确地向企业发送预警信息。
发明内容
鉴于以上内容,有必要提供一种企业舆情分析方法、装置、电子设备及介质,能够充分利用舆情信息,以及时、准确地向企业发送预警信息。
本申请的第一方面提供一种企业舆情分析方法,所述企业舆情分析方法包括:
从预设的数据渠道获取舆情文本;
对获取到的舆情文本进行预处理,得到每个舆情文本的多个词组;
对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本;
基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果,并按照预设的分类标准对所述目标文本进行分类,得到分类结果;
将所述目标文本、所述识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果;
选取出情感结果为负面情绪的目标文本,作为预警文本;
将所述预警文本与配置列表中的预警词组进行匹配,并将匹配到的预警词组确定为所述预警文本的预警类别;
获取所述预警文本的目标企业,并根据所述预警类别及所述预警文本生成所述目标企业的预警信息。
本申请的第二方面提供一种电子设备,所述电子设备包括处理器和存储器,所述处理器用于执行所述存储器中存储的计算机可读指令以实现以下步骤:
从预设的数据渠道获取舆情文本;
对获取到的舆情文本进行预处理,得到每个舆情文本的多个词组;
对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本;
基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果,并按照预设的分类标准对所述目标文本进行分类,得到分类结果;
将所述目标文本、所述识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果;
选取出情感结果为负面情绪的目标文本,作为预警文本;
将所述预警文本与配置列表中的预警词组进行匹配,并将匹配到的预警词组确定为所述预警文本的预警类别;
获取所述预警文本的目标企业,并根据所述预警类别及所述预警文本生成所述目标企业的预警信息。
本申请的第三方面提供一种计算机可读存储介质,所述计算机可读存储介质上存储有至少一个计算机可读指令,所述至少一个计算机可读指令被处理器执行以实现以下步骤:
从预设的数据渠道获取舆情文本;
对获取到的舆情文本进行预处理,得到每个舆情文本的多个词组;
对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本;
基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果,并按照预设的分类标准对所述目标文本进行分类,得到分类结果;
将所述目标文本、所述识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果;
选取出情感结果为负面情绪的目标文本,作为预警文本;
将所述预警文本与配置列表中的预警词组进行匹配,并将匹配到的预警词组确定为所述预警文本的预警类别;
获取所述预警文本的目标企业,并根据所述预警类别及所述预警文本生成所述目标企业的预警信息。
本申请的第四方面提供一种企业舆情分析装置,所述企业舆情分析装置包括:
获取单元,用于从预设的数据渠道获取舆情文本;
预处理单元,用于对获取到的舆情文本进行预处理,得到每个舆情文本的多个词组;
确定单元,用于对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本;
分类单元,用于基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果,并按照预设的分类标准对所述目标文本进行分类,得到分类结果;
输入单元,用于将所述目标文本、所述识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果;
选取单元,用于选取出情感结果为负面情绪的目标文本,作为预警文本;
所述确定单元,还用于将所述预警文本与配置列表中的预警词组进行匹配,并将匹配到的预警词组确定为所述预警文本的预警类别;
生成单元,用于获取所述预警文本的目标企业,并根据所述预警类别及所述预警文本生成所述目标企业的预警信息。
由以上技术方案可知,本申请应用人工智能,能够从海量数据中获取舆情文本进行自然语言处理,通过充分利用舆情信息以及时、准确地向企业发送预警信息。此外,本申请还应用于智慧医疗等智慧城市领域,通过分析疫情等舆情信息确定应对政策,从而推动智慧城市的发展。
附图说明
图1是本申请公开的一种企业舆情分析方法的较佳实施例的流程图。
图2是本申请公开的一种企业舆情分析装置的较佳实施例的功能模块图。
图3是本申请实现企业舆情分析方法的较佳实施例的电子设备的结构示意图。
具体实施方式
本申请实施例的企业舆情分析方法应用在电子设备中,也可以应用在电子设备和通过网络与所述电子设备进行连接的服务器所构成的硬件环境中,由服务器和电子设备共同执行。网络包括但不限于:广域网、城域网或局域网。
请参见图1,图1是本申请公开的一种企业舆情分析方法的较佳实施例的流程图。其中,根据不同的需求,该流程图中步骤的顺序可以改变,某些步骤可以省略。
S10,从预设的数据渠道获取舆情文本。
在本申请的至少一个实施例中,所述预设的数据渠道包括,但不仅限于:报纸、微博、微信、论坛等,从上述数据渠道获取到的舆情文本主要包括新闻、论坛帖子、微博博文、微信文章等。
在本申请的至少一个实施例中,所述电子设备从预设的数据渠道获取舆情文本包括:
(1)所述电子设备利用光学字符识别技术(Optical Character Recognition,OCR)对纸质文本进行扫描识别,将扫描到的电子文本作为所述舆情文本。
(2)所述电子设备基于网络爬虫技术,从社交软件的开放接口获取所述社交软件中用户发布的文本,作为所述舆情文本。
(3)所述电子设备通过匿名代理池获取门户网站中用户发布的文本,作为所述舆情文本。
具体地,所述电子设备通过匿名代理池获取门户网站中用户发布的文本包括:
基于普通HTTP(HyperText Transfer Protocol,超文本传输协议)代理及高匿名代理构成所述匿名代理池,所述电子设备从所述普通HTTP代理及所述高匿名代理生成的地址中随机选取可变地址,进一步地,所述电子设备将预设头文件与所述可变地址进行拼接,将拼接后的地址作为爬虫从所述门户网站中获取用户发布的文本。
通过所述预设头文件与所述可变地址进行拼接,可以降低所述爬虫被所述门户网站识别的概率。
S11,对获取到的舆情文本进行预处理,得到每个舆情文本的多个词组。
在本申请的至少一个实施例中,所述电子设备对获取到的舆情文本进行预处理,得到每个舆情文本对应的多个词组包括:
所述电子设备过滤所述舆情文本中的特殊字符及停用词,得到第一文本,所述电子设备基于余弦距离公式,对所述第一文本进行去重处理,得到第二文本,所述电子设备根据预设的自定义词典对所述第二文本进行切分,得到切分位置,并根据所述切分位置,构建至少一个有向无环图,所述电子设备根据所述自定义词典中的权值计算每个有向无环图的概率,并将概率最大的有向无环图对应的切分位置确定为目标切分位置,所述电子设备根据所述目标切分位置确定每个舆情文本的多个词组。
其中,所述特殊字符包括,但不限于:表情符号、符号图案等。
进一步地,所述预设的自定义词典中存储至少一个自定义词及每个自定义词对应的权值。
通过上述实施方式,能够去除所述第一文本中重复的文本,避免处理重复文本所耗费的时间;通过将概率最大的切分位置确定为目标切分位置,能够准确地切分舆情文本。
具体地,所述电子设备基于余弦距离公式,对所述第一文本进行去重处理,得到第二文本包括:
所述电子设备根据所述第一文本的标题,计算所述第一文本的哈希值,所述电子设备从所述第一文本中抽取预设特征并建立特征索引,并根据所述第一文本的哈希值,采用余弦距离公式计算所述第一文本中任意两个第一文本的相似距离,得到文本对的相似距离,其中, 所述文本对包括任意两个第一文本,所述电子设备通过所述特征索引搜索出相似距离大于预设值的文本对,并将该文本对确定为相似文本对,所述电子设备判断所述相似文本对中的预设特征是否相同,当所述相似文本对中的预设特征相同时,所述电子设备剔除所述相似文本对中的任意一个第一文本,并将保留的第一文本确定为所述第二文本。
通过上述实施方式,能够准确快速剔除重复文本。
S12,对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本。
在本申请的至少一个实施例中,所述电子设备对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本包括:
所述电子设备对每个舆情文本的多个词组进行向量化处理,得到每个舆情文本的输入向量序列,进一步地,所述电子设备将每个舆情文本的输入向量序列输入到NER模型中,并获取激活层中每个序列位置上对应的每个标签的输出概率及转移概率,对于每个序列位置,所述电子设备对每个标签的输出概率及转移概率进行加权和运算,得到每个标签的数值,所述电子设备将数值最高的标签确定为每个序列位置上的输出标签,并组合每个序列位置上的输出标签,得到每个舆情文本的实体列表,当检测到实体列表中含有一个企业实体时,所述电子设备将该实体列表对应的舆情文本确定为所述目标文本,或者当检测到实体列表中含有多个企业实体时,所述电子设备将该实体列表对应的舆情文本确定为多文本,依照所述多个企业实体拆分所述多文本,得到所述目标文本。
例如:对于每个输入向量序列X=(x 1,x 2,…,x n),将X=(x 1,x 2,…,x n)输入NER模型中,所述电子设备从激活层能够获取到第i个序列位置上每个标签的输出概率,分别为:P iy1,P iy2,…,P iym;所述电子设备从激活层能够获取到第i个序列位置上每个标签的转移概率,分别为:A iy,A iy2,…,A iym。经计算,得到第i个序列位置上每个标签(如,标签B-PER)的数值,即:S iy,S iy2,…,S iym,若S iy2的数值最大,将S iy2对应的标签确定为第i个序列位置上的输出标签,依次即可得到实体列表Y=(y 1,y 2,…,y n),在实体列表中的表示形式中,B-PER表示人名的首字符标记,E-PER表示人名的结尾字符标记,O表示独立字符标记,B-COM表示企业名称的首字符标记,I-COM表示企业名称的中间字符标记,E-COM表示企业名称的结尾字符标记,实体列表A的表示形式为(B-PER,E-PER,O,O,B-COM,I-COM,E-COM),所述电子设备确定所述实体列表A中含有一个企业实体,该企业实体为(B-COM,I-COM,E-COM),实体列表B的表示形式为(B-COM1,E-COM1,O,O,B-COM2,I-COM2,E-COM2),所述电子设备确定所述实体列表B中含有两个企业实体,两个企业实体分别为(B-COM1,E-COM1)及(B-COM2,I-COM2,E-COM2)。
承接上面的例子,(B-COM1,E-COM1)对应的企业实体可以为“T企”;(B-COM2,I-COM2,E-COM2)对应的企业实体可以为“G公司”。
通过上述实施方式,能够准确地确定出具有企业实体的舆情文本,避免后续主体识别时无法确定出企业实体,进而能够对多余舆情的文本进行处理。
具体地,在将每个舆情文本的输入向量序列输入到NER模型之前,所述方法还包括:
所述电子设备获取企业舆情数据,并根据所述企业舆情数据选取目标标注模式,所述电子设备将所述目标标注模式添加至组合模型中,得到标注模型,所述电子设备对所述标注模型进行裁剪,获得裁剪模型,所述电子设备对所述裁剪模型进行降阶,获得所述NER模型。
其中,所述组合模型包括:
(1)基于长短期记忆网络(Long Short-Term Memory,LSTM)和条件随机场(Conditional Random Fields,CRF)的模型。
(2)基于双向长短期记忆网络(Bi LSTM)和条件随机场的模型。
(3)基于BiGRU和条件随机场的模型。
通过训练组合模型,能够得到更准确的识别效果,进而通过对标注模型进行相对熵裁剪,获得相对于标注模型较小的裁剪模型,在基于相对熵裁剪的基础上,同时降低模型的阶数,对模型网络中的维度进行降阶,从而可以减少最终解码网络的复杂度,提高了命名实体识别的速度。
具体地,所述电子设备对所述标注模型进行裁剪,获得裁剪模型包括:
所述电子设备从所述标注模型中提取所有卷积核,进一步地,所述电子设备利用灰色关联分析方法对所述所有卷积核中每个卷积核进行重要度的量化,得到每个卷积核的重要度的量化值,并将所述所有卷积核按照所述量化值的大小依照从小至大的顺序进行排序,得到队列,所述电子设备从所述队列中选取前N个卷积核,作为目标卷积核,所述N为正整数,所述电子设备删除所述标注模型中的目标卷积核,得到所述裁剪模型。
通过上述实施方式,能够在保证模型精度的情况下,实现对模型的裁剪,进而能够提高命名实体识别效率。
在本申请的至少一个实施例中,所述电子设备对每个舆情文本的多个词组进行向量化处理,得到每个舆情文本的输入向量序列包括:
所述电子设备根据预设编码表获取每个舆情文本的每个词组的编码向量,并根据每个舆情文本的每个词组的位置编号生成每个词组的位置向量,所述电子设备拼接每个词组的编码向量及每个词组的位置向量,得到每个词组的目标向量,所述电子设备依照词序组合每个舆情文本的每个词组的目标向量,得到每个舆情文本的输入向量序列。
通过结合每个词组在文本中的位置编号而生成每个词组的目标向量,使生成的目标向量具有上下文语义特征,进而能够提高词组命名实体识别的准确性。
在本申请的一个实施例中,为进一步保证舆情文本的私密和安全性,本实施例还可以将舆情文本存储于一区块链的节点中。
在本申请的至少一个实施例中,当检测到实体列表中含有多个企业实体时,将该实体列表对应的舆情文本确定为多文本,依照所述多个企业实体拆分所述多文本,得到所述目标文本。
例如:舆情文本1为:A企业是一家知名度很高的企业。B企业的销售量在行业内领先。当检测到实体列表中含有A企业实体及B企业实体时,所述电子设备将所述舆情文本1确定为多文本,依据A企业实体及B企业实体拆分所述舆情文本1,得到目标文本2为:A企业是一家知名度很高的企业;目标文本3为:B企业的销售量在行业内领先。
S13,基于依存句法分析技术,对所述目标文本进行主体判别主体识别,得到判别结果识别结果,并按照预设的分类标准对所述目标文本进行分类,得到分类结果。
在本申请的至少一个实施例中,所述判别结果识别结果包括:企业在目标文本中是主体,或者企业在目标文本中是提及。
所述分类结果包括:招聘类、广告类、咨询类等。
在其他实施例中,所述电子设备可以从多个维度上对所述目标文本进行分类,每个维度上有多个分类结果。
在本申请的至少一个实施例中,所述电子设备基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果包括:
所述电子设备根据依存句法分析技术,获取所述目标文本中每个文本语句的核心动词,所述电子设备确定与所述核心动词依存关系为主谓关系和动宾关系的分词,所述电子设备计算所述目标文本中所述分词的总数量,所述电子设备获取所述企业实体对应的目标分词,并计算所述目标分词在所述目标文本中的目标数量,所述电子设备将所述目标数量除以所述总数量,得到目标比例,当检测到所述目标比例大于第一预设比例时,所述电子设备将所述目标分词确定为所述目标文本的主体,作为所述识别结果,或者当检测到所述目标比例小于第 二预设比例时,所述电子设备将所述目标分词确定为所述目标文本的提及,作为所述识别结果。
例如:核心动词为“做出”,然后根据依存句法分析技术寻找与核心动词依存关系为“主谓关系”和“动宾关系”的词,分别为“A企业”和“通报”。
需要说明的是,当企业在目标文本中是主体时,企业相应的目标分词在目标文本中所占的比例大于所述第一预设比例,当企业在目标文本中是提及时,企业相应的目标分词在目标文本中所占的比例小于所述第二预设比例。
通过依存句法分析技术能够快速确定核心动词及分词,进而通过目标分词所占的比例的检测,能够准确确定识别结果。
S14,将所述目标文本、所述判别结果识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果。
在本申请的至少一个实施例中,所述情感结果包括:中立情绪、负面情绪、正面情绪。
在本申请的至少一个实施例中,在将所述目标文本、所述判别结果识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果之前,所述方法还包括:
所述电子设备采用爬虫程序获取所有分类结果的第一历史数据,所述电子设备将所述第一历史数据输入到遗忘门层进行遗忘处理,得到训练数据,所述电子设备采用交叉验证法将所述训练数据划分为训练集及验证集,所述电子设备将所述训练集中的数据输入到输入门层进行训练,得到初级学习器,所述电子设备根据所述验证集中的数据,调整所述初级学习器,得到次级学习器,所述电子设备根据所述目标文本的分类结果获取第二历史数据,并将所述第二历史数据作为测试数据测试所述次级学习器,得到测试结果,所述电子设备计算通过测试的第二历史数据的目标数量,及计算参与测试的第二历史数据的总数量,所述电子设备将所述目标数量除以所述总数量,得到测试成功率,当所述测试成功率大于配置值时,所述电子设备将所述次级学习器确定为所述情感模型,或者当所述测试成功率小于或者等于所述配置值时,所述电子设备根据所述第二历史数据调整所述次级学习器,得到所述情感模型。
通过上述实施方式,能够使训练到的所述情感模型能够更加准确。
S15,选取出情感结果为负面情绪的目标文本,作为预警文本。
在本申请的至少一个实施例中,当舆情文本的情感结果为中立情绪或者正面情绪时,所述电子设备无需对该舆情文本进行预警。
S16,将所述预警文本与配置列表中的预警词组进行匹配,并将匹配到的预警词组确定为所述预警文本的预警类别。
在本申请的至少一个实施例中,所述配置列表中存储多个预警词组,所述预警词组包括:法律诉讼、人事变动及警方处罚等。
S17,获取所述预警文本的目标企业,并根据所述预警类别及所述预警文本生成所述目标企业的预警信息。
在本申请的至少一个实施例中,所述电子设备根据所述预警类别及所述预警文本生成所述目标企业的预警信息包括:
所述电子设备根据所述预警类别从案例库中提取策略,进一步地,所述电子设备根据所述预警类别、所述预警文本及所述策略生成所述预警信息。
需要强调的是,为进一步保证上述预警信息的私密和安全性,上述预警信息还可以存储于一区块链的节点中。
在图1所描述的方法流程中,能够从海量数据中获取舆情文本进行自然语言处理,通过充分利用舆情信息以及时、准确地向企业发送预警信息。此外,本申请应用人工智能及智慧医疗等智慧城市领域,通过分析疫情等舆情信息确定应对政策,从而推动智慧城市的发展。
以上所述,仅是本申请的具体实施方式,但本申请的保护范围并不局限于此,对于本领 域的普通技术人员来说,在不脱离本申请创造构思的前提下,还可以做出改进,但这些均属于本申请的保护范围。
请参见图2,图2是本申请公开的一种企业舆情分析装置的较佳实施例的功能模块图。
在一些实施例中,所述企业舆情分析装置运行于电子设备中。所述企业舆情分析装置可以包括多个由程序代码段所组成的功能模块,所述程序是一系列的计算机可读指令代码。所述企业舆情分析装置中的各个程序段的程序代码可以存储于存储器中,并由至少一个处理器所执行,以执行图1所描述的企业舆情分析方法中的部分或全部步骤,具体可以参照图1所述方法中的相关描述,在此不再赘述。
本实施例中,所述企业舆情分析装置根据其所执行的功能,可以被划分为多个功能模块。所述功能模块可以包括:获取单元110、预处理单元111、确定单元112、分类单元113、输入单元114、选取单元115、生成单元116、添加单元117、裁剪单元118、降阶单元119、划分单元120、调整单元121、测试单元122及计算单元123。本申请所称的模块是指一种能够被至少一个处理器所执行并且能够完成固定功能的一系列计算机可读指令段,其存储在存储器中。
获取单元110从预设的数据渠道获取舆情文本。
在本申请的至少一个实施例中,所述预设的数据渠道包括,但不仅限于:报纸、微博、微信、论坛等,从上述数据渠道获取到的舆情文本主要包括新闻、论坛帖子、微博博文、微信文章等。
在本申请的至少一个实施例中,所述获取单元110从预设的数据渠道获取舆情文本包括:
(1)所述获取单元110利用光学字符识别技术(Optical Character Recognition,OCR)对纸质文本进行扫描识别,将扫描到的电子文本作为所述舆情文本。
(2)所述获取单元110基于网络爬虫技术,从社交软件的开放接口获取所述社交软件中用户发布的文本,作为所述舆情文本。
(3)所述获取单元110通过匿名代理池获取门户网站中用户发布的文本,作为所述舆情文本。
具体地,所述获取单元110通过匿名代理池获取门户网站中用户发布的文本包括:
基于普通HTTP(HyperText Transfer Protocol,超文本传输协议)代理及高匿名代理构成所述匿名代理池,所述获取单元110从所述普通HTTP代理及所述高匿名代理生成的地址中随机选取可变地址,进一步地,所述获取单元110将预设头文件与所述可变地址进行拼接,将拼接后的地址作为爬虫从所述门户网站中获取用户发布的文本。
通过所述预设头文件与所述可变地址进行拼接,可以降低所述爬虫被所述门户网站识别的概率。
预处理单元111对获取到的舆情文本进行预处理,得到每个舆情文本的多个词组。
在本申请的至少一个实施例中,所述预处理单元111对获取到的舆情文本进行预处理,得到每个舆情文本对应的多个词组包括:
所述预处理单元111过滤所述舆情文本中的特殊字符及停用词,得到第一文本,所述预处理单元111基于余弦距离公式,对所述第一文本进行去重处理,得到第二文本,所述预处理单元111根据预设的自定义词典对所述第二文本进行切分,得到切分位置,并根据所述切分位置,构建至少一个有向无环图,所述预处理单元111根据所述自定义词典中的权值计算每个有向无环图的概率,并将概率最大的有向无环图对应的切分位置确定为目标切分位置,所述预处理单元111根据所述目标切分位置确定每个舆情文本的多个词组。
其中,所述特殊字符包括,但不限于:表情符号、符号图案等。
进一步地,所述预设的自定义词典中存储至少一个自定义词及每个自定义词对应的权值。
通过上述实施方式,能够去除所述第一文本中重复的文本,避免处理重复文本所耗费的 时间;通过将概率最大的切分位置确定为目标切分位置,能够准确地切分舆情文本。
具体地,所述预处理单元111基于余弦距离公式,对所述第一文本进行去重处理,得到第二文本包括:
所述预处理单元111根据所述第一文本的标题,计算所述第一文本的哈希值,所述预处理单元111从所述第一文本中抽取预设特征并建立特征索引,并根据所述第一文本的哈希值,采用余弦距离公式计算所述第一文本中任意两个第一文本的相似距离,得到文本对的相似距离,其中,所述文本对包括任意两个第一文本,所述预处理单元111通过所述特征索引搜索出相似距离大于预设值的文本对,并将该文本对确定为相似文本对,所述预处理单元111判断所述相似文本对中的预设特征是否相同,当所述相似文本对中的预设特征相同时,所述预处理单元111剔除所述相似文本对中的任意一个第一文本,并将保留的第一文本确定为所述第二文本。
通过上述实施方式,能够准确快速剔除重复文本。
确定单元112对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本。
在本申请的至少一个实施例中,所述确定单元112对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本包括:
所述确定单元112对每个舆情文本的多个词组进行向量化处理,得到每个舆情文本的输入向量序列,进一步地,所述确定单元112将每个舆情文本的输入向量序列输入到NER模型中,并获取激活层中每个序列位置上对应的每个标签的输出概率及转移概率,对于每个序列位置,所述确定单元112对每个标签的输出概率及转移概率进行加权和运算,得到每个标签的数值,所述确定单元112将数值最高的标签确定为每个序列位置上的输出标签,并组合每个序列位置上的输出标签,得到每个舆情文本的实体列表,当检测到实体列表中含有一个企业实体时,所述确定单元112将该实体列表对应的舆情文本确定为所述目标文本,或者当检测到实体列表中含有多个企业实体时,所述确定单元112将该实体列表对应的舆情文本确定为多文本,依照所述多个企业实体拆分所述多文本,得到所述目标文本。
例如:对于每个输入向量序列X=(x 1,x 2,…,x n),将X=(x 1,x 2,…,x n)输入NER模型中,所述确定单元112从激活层能够获取到第i个序列位置上每个标签的输出概率,分别为:P iy,P iy,…,P iym;所述确定单元112从激活层能够获取到第i个序列位置上每个标签的转移概率,分别为:A iy1,A iy2,…,A iym。经计算,得到第i个序列位置上每个标签(如,标签B-PER)的数值,即:S iy1,S iy2,…,S iym,若S iy2的数值最大,将S iy2对应的标签确定为第i个序列位置上的输出标签,依次即可得到实体列表Y=(y 1,y 2,…,y n),在实体列表中的表示形式中,B-PER表示人名的首字符标记,E-PER表示人名的结尾字符标记,O表示独立字符标记,B-COM表示企业名称的首字符标记,I-COM表示企业名称的中间字符标记,E-COM表示企业名称的结尾字符标记,实体列表A的表示形式为(B-PER,E-PER,O,O,B-COM,I-COM,E-COM),所述电子设备确定所述实体列表A中含有一个企业实体,该企业实体为(B-COM,I-COM,E-COM),实体列表B的表示形式为(B-COM1,E-COM1,O,O,B-COM2,I-COM2,E-COM2),所述确定单元112确定所述实体列表B中含有两个企业实体,两个企业实体分别为(B-COM1,E-COM1)及(B-COM2,I-COM2,E-COM2)。
承接上面的例子,(B-COM1,E-COM1)对应的企业实体可以为“T企”;(B-COM2,I-COM2,E-COM2)对应的企业实体可以为“G公司”。
通过上述实施方式,能够准确地确定出具有企业实体的舆情文本,避免后续主体识别时无法确定出企业实体,进而能够对多余舆情的文本进行处理。
具体地,在将每个舆情文本的输入向量序列输入到NER模型之前,所述获取单元110获取企业舆情数据,并根据所述企业舆情数据选取目标标注模式,添加单元117将所述目标标 注模式添加至组合模型中,得到标注模型,裁剪单元118对所述标注模型进行裁剪,获得裁剪模型,降阶单元119对所述裁剪模型进行降阶,获得所述NER模型。
其中,所述组合模型包括:
(1)基于长短期记忆网络(Long Short-Term Memory,LSTM)和条件随机场(Conditional Random Fields,CRF)的模型。
(2)基于双向长短期记忆网络(Bi LSTM)和条件随机场的模型。
(3)基于BiGRU和条件随机场的模型。
通过训练组合模型,能够得到更准确的识别效果,进而通过对标注模型进行相对熵裁剪,获得相对于标注模型较小的裁剪模型,在基于相对熵裁剪的基础上,同时降低模型的阶数,对模型网络中的维度进行降阶,从而可以减少最终解码网络的复杂度,提高了命名实体识别的速度。
具体地,所述裁剪单元118对所述标注模型进行裁剪,获得裁剪模型包括:
所述裁剪单元118从所述标注模型中提取所有卷积核,进一步地,所述裁剪单元118利用灰色关联分析方法对所述所有卷积核中每个卷积核进行重要度的量化,得到每个卷积核的重要度的量化值,并将所述所有卷积核按照所述量化值的大小依照从小至大的顺序进行排序,得到队列,所述裁剪单元118从所述队列中选取前N个卷积核,作为目标卷积核,所述N为正整数,所述裁剪单元118删除所述标注模型中的目标卷积核,得到所述裁剪模型。
通过上述实施方式,能够在保证模型精度的情况下,实现对模型的裁剪,进而能够提高命名实体识别效率。
在本申请的至少一个实施例中,所述确定单元112对每个舆情文本的多个词组进行向量化处理,得到每个舆情文本的输入向量序列包括:
所述确定单元112根据预设编码表获取每个舆情文本的每个词组的编码向量,并根据每个舆情文本的每个词组的位置编号生成每个词组的位置向量,所述确定单元112拼接每个词组的编码向量及每个词组的位置向量,得到每个词组的目标向量,所述确定单元112依照词序组合每个舆情文本的每个词组的目标向量,得到每个舆情文本的输入向量序列。
通过结合每个词组在文本中的位置编号而生成每个词组的目标向量,使生成的目标向量具有上下文语义特征,进而能够提高词组命名实体识别的准确性。
在本申请的至少一个实施例中,当检测到实体列表中含有多个企业实体时,将该实体列表对应的舆情文本确定为多文本,依照所述多个企业实体拆分所述多文本,得到所述目标文本。
例如:舆情文本1为:A企业是一家知名度很高的企业。B企业的销售量在行业内领先。当检测到实体列表中含有A企业实体及B企业实体时,所述电子设备将所述舆情文本1确定为多文本,依据A企业实体及B企业实体拆分所述舆情文本1,得到目标文本2为:A企业是一家知名度很高的企业;目标文本3为:B企业的销售量在行业内领先。
分类单元113基于依存句法分析技术,对所述目标文本进行主体判别主体识别,得到判别结果识别结果,并按照预设的分类标准对所述目标文本进行分类,得到分类结果。
在本申请的至少一个实施例中,所述判别结果识别结果包括:企业在目标文本中是主体,或者企业在目标文本中是提及。
所述分类结果包括:招聘类、广告类、咨询类等。
在其他实施例中,所述分类单元113可以从多个维度上对所述目标文本进行分类,每个维度上有多个分类结果。
在本申请的至少一个实施例中,所述分类单元113基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果包括:
所述分类单元113根据依存句法分析技术,获取所述目标文本中每个文本语句的核心动 词,所述分类单元113确定与所述核心动词依存关系为主谓关系和动宾关系的分词,所述分类单元113计算所述目标文本中所述分词的总数量,所述分类单元113获取所述企业实体对应的目标分词,并计算所述目标分词在所述目标文本中的目标数量,所述分类单元113将所述目标数量除以所述总数量,得到目标比例,当检测到所述目标比例大于第一预设比例时,所述分类单元113将所述目标分词确定为所述目标文本的主体,作为所述识别结果,或者当检测到所述目标比例小于第二预设比例时,所述分类单元113将所述目标分词确定为所述目标文本的提及,作为所述识别结果。
例如:核心动词为“做出”,然后根据依存句法分析技术寻找与核心动词依存关系为“主谓关系”和“动宾关系”的词,分别为“A企业”和“通报”。
需要说明的是,当企业在目标文本中是主体时,企业相应的目标分词在目标文本中所占的比例大于所述第一预设比例,当企业在目标文本中是提及时,企业相应的目标分词在目标文本中所占的比例小于所述第二预设比例。
通过依存句法分析技术能够快速确定核心动词及分词,进而通过目标分词所占的比例的检测,能够准确确定识别结果。
输入单元114将所述目标文本、所述判别结果识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果。
在本申请的至少一个实施例中,所述情感结果包括:中立情绪、负面情绪、正面情绪。
在本申请的至少一个实施例中,在将所述目标文本、所述判别结果识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果之前,所述获取单元110采用爬虫程序获取所有分类结果的第一历史数据,所述输入单元114将所述第一历史数据输入到遗忘门层进行遗忘处理,得到训练数据,划分单元120采用交叉验证法将所述训练数据划分为训练集及验证集,所述输入单元114将所述训练集中的数据输入到输入门层进行训练,得到初级学习器,调整单元121根据所述验证集中的数据,调整所述初级学习器,得到次级学习器,所述获取单元110根据所述目标文本的分类结果获取第二历史数据,测试单元122将所述第二历史数据作为测试数据测试所述次级学习器,得到测试结果,计算单元123计算通过测试的第二历史数据的目标数量,及计算参与测试的第二历史数据的总数量,所述计算单元123将所述目标数量除以所述总数量,得到测试成功率,当所述测试成功率大于配置值时,所述确定单元112将所述次级学习器确定为所述情感模型,或者当所述测试成功率小于或者等于所述配置值时,所述调整单元121根据所述第二历史数据调整所述次级学习器,得到所述情感模型。
通过上述实施方式,能够使训练到的所述情感模型能够更加准确。
选取单元115选取出情感结果为负面情绪的目标文本,作为预警文本。
在本申请的至少一个实施例中,当舆情文本的情感结果为中立情绪或者正面情绪时,无需对该舆情文本进行预警。
所述确定单元112将所述预警文本与配置列表中的预警词组进行匹配,并将匹配到的预警词组确定为所述预警文本的预警类别。
在本申请的至少一个实施例中,所述配置列表中存储多个预警词组,所述预警词组包括:法律诉讼、人事变动及警方处罚等。
生成单元116获取所述预警文本的目标企业,并根据所述预警类别及所述预警文本生成所述目标企业的预警信息。
在本申请的至少一个实施例中,所述生成单元116根据所述预警类别及所述预警文本生成所述目标企业的预警信息包括:
所述生成单元116根据所述预警类别从案例库中提取策略,进一步地,所述生成单元116根据所述预警类别、所述预警文本及所述策略生成所述预警信息。
需要强调的是,为进一步保证上述预警信息的私密和安全性,上述预警信息还可以存储于一区块链的节点中。
在图2所描述的企业舆情分析装置中,能够从海量数据中获取舆情文本进行自然语言处理,通过充分利用舆情信息以及时、准确地向企业发送预警信息。此外,本申请应用人工智能及智慧医疗等智慧城市领域,通过分析疫情等舆情信息确定应对政策,从而推动智慧城市的发展。
如图3所示,图3是本申请实现企业舆情分析方法的较佳实施例的电子设备的结构示意图。所述电子设备3包括存储器31、至少一个处理器32、存储在所述存储器31中并可在所述至少一个处理器32上运行的计算机可读指令33及至少一条通讯总线34。
本领域技术人员可以理解,图3所示的示意图仅仅是所述电子设备3的示例,并不构成对所述电子设备3的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件,例如所述电子设备3还可以包括输入输出设备、网络接入设备等。
所述电子设备3还包括但不限于任何一种可与用户通过键盘、鼠标、遥控器、触摸板或声控设备等方式进行人机交互的电子产品,例如,个人计算机、平板电脑、智能手机、个人数字助理(Personal Digital Assistant,PDA)、游戏机、交互式网络电视(Internet Protocol Television,IPTV)、智能式穿戴式设备等。
所述至少一个处理器32可以是中央处理单元(Central Processing Unit,CPU),还可以是其他通用处理器、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、分立门或者晶体管逻辑器件、分立硬件组件等。该处理器32可以是微处理器或者该处理器32也可以是任何常规的处理器等,所述处理器32是所述电子设备3的控制中心,利用各种接口和线路连接整个电子设备3的各个部分。
所述存储器31可用于存储所述计算机可读指令33和/或模块/单元,所述处理器32通过运行或执行存储在所述存储器31内的计算机可读指令和/或模块/单元,以及调用存储在存储器31内的数据,实现所述电子设备3的各种功能。所述存储器31可主要包括存储程序区和存储数据区,其中,存储程序区可存储操作系统、至少一个功能所需的应用程序(比如声音播放功能、图像播放功能等)等;存储数据区可存储根据电子设备3的使用所创建的数据(比如音频数据)等。此外,存储器31可以包括高速随机存取存储器等易失性存储器,还可以包括非易失性存储器,例如硬盘、内存、插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)、至少一个磁盘存储器件、闪存器件、或其他非易失性固态存储器件。
结合图1,所述电子设备3中的所述存储器31存储多个指令以实现一种企业舆情分析方法,所述处理器32可执行所述多个指令从而实现:从预设的数据渠道获取舆情文本;对获取到的舆情文本进行预处理,得到每个舆情文本的多个词组;对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本;基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果,并按照预设的分类标准对所述目标文本进行分类,得到分类结果;将所述目标文本、所述识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果;选取出情感结果为负面情绪的目标文本,作为预警文本;将所述预警文本与配置列表中的预警词组进行匹配,并将匹配到的预警词组确定为所述预警文本的预警类别;获取所述预警文本的目标企业,并根据所述预警类别及所述预警文本生成所述目标企业的预警信息。
具体地,所述处理器32对上述指令的具体实现方法可参考图1对应实施例中相关步骤的描述,在此不赘述。
在图3所描述的电子设备3中,能够从海量数据中获取舆情文本进行自然语言处理,通过充分利用舆情信息以及时、准确地向企业发送预警信息。此外,本申请应用人工智能及智慧医疗等智慧城市领域,通过分析疫情等舆情信息确定应对政策,从而推动智慧城市的发展。
所述电子设备3集成的模块/单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中,所述计算机可读存储介质可以是非易失性的存储介质,也可以是易失性的存储介质。基于这样的理解,本申请实现上述实施例方法中的全部或部分流程,也可以通过计算机可读指令来指令相关的硬件来完成,所述的计算机可读指令可存储于计算机可读存储介质中,该计算机可读指令在被处理器执行时,可实现上述各个方法实施例的步骤。其中,所述计算机可读指令包括计算机可读指令代码,所述计算机可读指令代码可以为源代码形式、对象代码形式、可执行文件或某些中间形式等。所述计算机可读介质可以包括:能够携带所述计算机可读指令代码的任何实体或装置、记录介质、U盘、移动硬盘、磁碟、光盘、计算机存储器、只读存储器(ROM,Read-Only Memory)、随机存储器(RAM,Random Access Memory)。
在本申请所提供的几个实施例中,应该理解到,所揭露的系统,装置和方法,可以通过其它的方式实现。例如,以上所描述的装置实施例仅仅是示意性的,例如,所述模块的划分,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。
所述作为分离部件说明的模块可以是或者也可以不是物理上分开的,作为模块显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。
另外,在本申请各个实施例中的各功能模块可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用硬件加软件功能模块的形式实现。
对于本领域技术人员而言,显然本申请不限于上述示范性实施例的细节,而且在不背离本申请的精神或基本特征的情况下,能够以其他的具体形式实现本申请。因此,无论从哪一点来看,均应将实施例看作是示范性的,而且是非限制性的,本申请的范围由所附权利要求而不是上述说明限定,因此旨在将落在权利要求的等同要件的含义和范围内的所有变化涵括在本申请内。不应将权利要求中的任何附关联图标记视为限制所涉及的权利要求。此外,显然“包括”一词不排除其他单元或步骤,单数不排除复数。系统权利要求中陈述的多个单元或装置也可以由一个单元或装置通过软件或者硬件来实现。第二等词语用来表示名称,而并不表示任何特定的顺序。
最后应说明的是,以上实施例仅用以说明本申请的技术方案而非限制,尽管参照较佳实施例对本申请进行了详细说明,本领域的普通技术人员应当理解,可以对本申请的技术方案进行修改或等同替换,而不脱离本申请技术方案的精神和范围。

Claims (20)

  1. 一种企业舆情分析方法,其中,所述企业舆情分析方法包括:
    从预设的数据渠道获取舆情文本;
    对获取到的舆情文本进行预处理,得到每个舆情文本的多个词组;
    对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本;
    基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果,并按照预设的分类标准对所述目标文本进行分类,得到分类结果;
    将所述目标文本、所述识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果;
    选取出情感结果为负面情绪的目标文本,作为预警文本;
    将所述预警文本与配置列表中的预警词组进行匹配,并将匹配到的预警词组确定为所述预警文本的预警类别;
    获取所述预警文本的目标企业,并根据所述预警类别及所述预警文本生成所述目标企业的预警信息。
  2. 根据权利要求1所述的企业舆情分析方法,其中,所述对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本包括:
    对每个舆情文本的多个词组进行向量化处理,得到每个舆情文本的输入向量序列;
    将每个舆情文本的输入向量序列输入到NER模型中,并获取激活层中每个序列位置上对应的每个标签的输出概率及转移概率;
    对于每个序列位置,对每个标签的输出概率及转移概率进行加权和运算,得到每个标签的数值;
    将数值最高的标签确定为每个序列位置上的输出标签,并组合每个序列位置上的输出标签,得到每个舆情文本的实体列表;
    当检测到实体列表中含有一个企业实体时,将该实体列表对应的舆情文本确定为所述目标文本;或者
    当检测到实体列表中含有多个企业实体时,将该实体列表对应的舆情文本确定为多文本,依照所述多个企业实体拆分所述多文本,得到所述目标文本。
  3. 根据权利要求2所述的企业舆情分析方法,其中,所述对每个舆情文本的多个词组进行向量化处理,得到每个舆情文本的输入向量序列包括:
    根据预设编码表获取每个舆情文本的每个词组的编码向量;
    根据每个舆情文本的每个词组的位置编号生成每个词组的位置向量;
    拼接每个词组的编码向量及每个词组的位置向量,得到每个词组的目标向量;
    依照词序组合每个舆情文本的每个词组的目标向量,得到每个舆情文本的输入向量序列。
  4. 根据权利要求2所述的企业舆情分析方法,其中,在将每个舆情文本的输入向量序列输入到NER模型之前,所述方法还包括:
    获取企业舆情数据,并根据所述企业舆情数据选取目标标注模式;
    将所述目标标注模式添加至组合模型中,得到标注模型;
    对所述标注模型进行裁剪,获得裁剪模型;
    对所述裁剪模型进行降阶,获得所述NER模型。
  5. 根据权利要求4所述的企业舆情分析方法,其中,所述对所述标注模型进行裁剪,获得裁剪模型包括:
    从所述标注模型中提取所有卷积核;
    利用灰色关联分析方法对所述所有卷积核中每个卷积核进行重要度的量化,得到每个卷积核的重要度的量化值;
    将所述所有卷积核按照所述量化值的大小依照从小至大的顺序进行排序,得到队列;
    从所述队列中选取前N个卷积核,作为目标卷积核,所述N为正整数;
    删除所述标注模型中的目标卷积核,得到所述裁剪模型。
  6. 根据权利要求1所述的企业舆情分析方法,其中,所述基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果包括:
    根据依存句法分析技术,获取所述目标文本中每个文本语句的核心动词;
    确定与所述核心动词依存关系为主谓关系和动宾关系的分词;
    计算所述目标文本中所述分词的总数量;
    获取所述企业实体对应的目标分词,并计算所述目标分词在所述目标文本中的目标数量;
    将所述目标数量除以所述总数量,得到目标比例;
    当检测到所述目标比例大于第一预设比例时,将所述目标分词确定为所述目标文本的主体,作为所述识别结果,或者当检测到所述目标比例小于第二预设比例时,将所述目标分词确定为所述目标文本的提及,作为所述识别结果。
  7. 一种电子设备,其中,所述电子设备包括处理器和存储器,所述处理器用于执行存储器中存储的至少一个计算机可读指令以实现以下步骤:
    从预设的数据渠道获取舆情文本;
    对获取到的舆情文本进行预处理,得到每个舆情文本的多个词组;
    对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本;
    基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果,并按照预设的分类标准对所述目标文本进行分类,得到分类结果;
    将所述目标文本、所述识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果;
    选取出情感结果为负面情绪的目标文本,作为预警文本;
    将所述预警文本与配置列表中的预警词组进行匹配,并将匹配到的预警词组确定为所述预警文本的预警类别;
    获取所述预警文本的目标企业,并根据所述预警类别及所述预警文本生成所述目标企业的预警信息。
  8. 根据权利要求7所述的电子设备,其中,在所述对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本时,所述处理器执行所述至少一个计算机可读指令以实现以下步骤:
    对每个舆情文本的多个词组进行向量化处理,得到每个舆情文本的输入向量序列;
    将每个舆情文本的输入向量序列输入到NER模型中,并获取激活层中每个序列位置上对应的每个标签的输出概率及转移概率;
    对于每个序列位置,对每个标签的输出概率及转移概率进行加权和运算,得到每个标签的数值;
    将数值最高的标签确定为每个序列位置上的输出标签,并组合每个序列位置上的输出标签,得到每个舆情文本的实体列表;
    当检测到实体列表中含有一个企业实体时,将该实体列表对应的舆情文本确定为所述目标文本;或者
    当检测到实体列表中含有多个企业实体时,将该实体列表对应的舆情文本确定为多文本,依照所述多个企业实体拆分所述多文本,得到所述目标文本。
  9. 根据权利要求8所述的电子设备,其中,在所述对每个舆情文本的多个词组进行向量化处理,得到每个舆情文本的输入向量序列时,所述处理器执行所述至少一个计算机可读指令以实现以下步骤:
    根据预设编码表获取每个舆情文本的每个词组的编码向量;
    根据每个舆情文本的每个词组的位置编号生成每个词组的位置向量;
    拼接每个词组的编码向量及每个词组的位置向量,得到每个词组的目标向量;
    依照词序组合每个舆情文本的每个词组的目标向量,得到每个舆情文本的输入向量序列。
  10. 根据权利要求8所述的电子设备,其中,在将每个舆情文本的输入向量序列输入到NER模型之前,所述处理器执行所述至少一个计算机可读指令还用以实现以下步骤:
    获取企业舆情数据,并根据所述企业舆情数据选取目标标注模式;
    将所述目标标注模式添加至组合模型中,得到标注模型;
    对所述标注模型进行裁剪,获得裁剪模型;
    对所述裁剪模型进行降阶,获得所述NER模型。
  11. 根据权利要求10所述的电子设备,其中,在所述对所述标注模型进行裁剪,获得裁剪模型时,所述处理器执行所述至少一个计算机可读指令以实现以下步骤:
    从所述标注模型中提取所有卷积核;
    利用灰色关联分析方法对所述所有卷积核中每个卷积核进行重要度的量化,得到每个卷积核的重要度的量化值;
    将所述所有卷积核按照所述量化值的大小依照从小至大的顺序进行排序,得到队列;
    从所述队列中选取前N个卷积核,作为目标卷积核,所述N为正整数;
    删除所述标注模型中的目标卷积核,得到所述裁剪模型。
  12. 根据权利要求7所述的电子设备,其中,在所述基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果时,所述处理器执行所述至少一个计算机可读指令以实现以下步骤:
    根据依存句法分析技术,获取所述目标文本中每个文本语句的核心动词;
    确定与所述核心动词依存关系为主谓关系和动宾关系的分词;
    计算所述目标文本中所述分词的总数量;
    获取所述企业实体对应的目标分词,并计算所述目标分词在所述目标文本中的目标数量;
    将所述目标数量除以所述总数量,得到目标比例;
    当检测到所述目标比例大于第一预设比例时,将所述目标分词确定为所述目标文本的主体,作为所述识别结果,或者当检测到所述目标比例小于第二预设比例时,将所述目标分词确定为所述目标文本的提及,作为所述识别结果。
  13. 根据权利要求7所述的电子设备,其中,在将所述目标文本、所述识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果之前,所述处理器执行所述至少一个计算机可读指令还用以实现以下步骤:
    采用爬虫程序获取所有分类结果的第一历史数据;
    将所述第一历史数据输入到遗忘门层进行遗忘处理,得到训练数据;
    采用交叉验证法将所述训练数据划分为训练集及验证集;
    将所述训练集中的数据输入到输入门层进行训练,得到初级学习器;
    根据所述验证集中的数据,调整所述初级学习器,得到次级学习器;
    根据所述目标文本的分类结果获取第二历史数据;
    将所述第二历史数据作为测试数据测试所述次级学习器,得到测试结果;
    计算通过测试的第二历史数据的目标数量,及计算参与测试的第二历史数据的总数量;
    将所述目标数量除以所述总数量,得到测试成功率;
    当所述测试成功率大于配置值时,将所述次级学习器确定为所述情感模型;或者
    当所述测试成功率小于或者等于所述配置值时,根据所述第二历史数据调整所述次级学习器,得到所述情感模型。
  14. 一种计算机可读存储介质,其中,所述计算机可读存储介质存储有至少一个计算机可读指令,所述至少一个计算机可读指令被处理器执行时实现以下步骤:
    从预设的数据渠道获取舆情文本;
    对获取到的舆情文本进行预处理,得到每个舆情文本的多个词组;
    对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本;
    基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果,并按照预设的分类标准对所述目标文本进行分类,得到分类结果;
    将所述目标文本、所述识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果;
    选取出情感结果为负面情绪的目标文本,作为预警文本;
    将所述预警文本与配置列表中的预警词组进行匹配,并将匹配到的预警词组确定为所述预警文本的预警类别;
    获取所述预警文本的目标企业,并根据所述预警类别及所述预警文本生成所述目标企业的预警信息。
  15. 根据权利要求14所述的存储介质,其中,在所述对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本时,所述至少一个计算机可读指令被处理器执行以实现以下步骤:
    对每个舆情文本的多个词组进行向量化处理,得到每个舆情文本的输入向量序列;
    将每个舆情文本的输入向量序列输入到NER模型中,并获取激活层中每个序列位置上对应的每个标签的输出概率及转移概率;
    对于每个序列位置,对每个标签的输出概率及转移概率进行加权和运算,得到每个标签的数值;
    将数值最高的标签确定为每个序列位置上的输出标签,并组合每个序列位置上的输出标签,得到每个舆情文本的实体列表;
    当检测到实体列表中含有一个企业实体时,将该实体列表对应的舆情文本确定为所述目标文本;或者
    当检测到实体列表中含有多个企业实体时,将该实体列表对应的舆情文本确定为多文本,依照所述多个企业实体拆分所述多文本,得到所述目标文本。
  16. 根据权利要求15所述的存储介质,其中,在所述对每个舆情文本的多个词组进行向量化处理,得到每个舆情文本的输入向量序列时,所述至少一个计算机可读指令被处理器执行以实现以下步骤:
    根据预设编码表获取每个舆情文本的每个词组的编码向量;
    根据每个舆情文本的每个词组的位置编号生成每个词组的位置向量;
    拼接每个词组的编码向量及每个词组的位置向量,得到每个词组的目标向量;
    依照词序组合每个舆情文本的每个词组的目标向量,得到每个舆情文本的输入向量序列。
  17. 根据权利要求15所述的存储介质,其中,在将每个舆情文本的输入向量序列输入到NER模型之前,所述至少一个计算机可读指令被处理器执行还用以实现以下步骤:
    获取企业舆情数据,并根据所述企业舆情数据选取目标标注模式;
    将所述目标标注模式添加至组合模型中,得到标注模型;
    对所述标注模型进行裁剪,获得裁剪模型;
    对所述裁剪模型进行降阶,获得所述NER模型。
  18. 根据权利要求17所述的存储介质,其中,在所述对所述标注模型进行裁剪,获得裁剪模型时,所述至少一个计算机可读指令被处理器执行时还用以实现以下步骤:
    从所述标注模型中提取所有卷积核;
    利用灰色关联分析方法对所述所有卷积核中每个卷积核进行重要度的量化,得到每个卷积核的重要度的量化值;
    将所述所有卷积核按照所述量化值的大小依照从小至大的顺序进行排序,得到队列;
    从所述队列中选取前N个卷积核,作为目标卷积核,所述N为正整数;
    删除所述标注模型中的目标卷积核,得到所述裁剪模型。
  19. 根据权利要求14所述的存储介质,其中,在所述基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果时,所述至少一个计算机可读指令被处理器执行以实现以下步骤:
    根据依存句法分析技术,获取所述目标文本中每个文本语句的核心动词;
    确定与所述核心动词依存关系为主谓关系和动宾关系的分词;
    计算所述目标文本中所述分词的总数量;
    获取所述企业实体对应的目标分词,并计算所述目标分词在所述目标文本中的目标数量;
    将所述目标数量除以所述总数量,得到目标比例;
    当检测到所述目标比例大于第一预设比例时,将所述目标分词确定为所述目标文本的主体,作为所述识别结果,或者当检测到所述目标比例小于第二预设比例时,将所述目标分词确定为所述目标文本的提及,作为所述识别结果。
  20. 一种企业舆情分析装置,其中,所述企业舆情分析装置包括:
    获取单元,用于从预设的数据渠道获取舆情文本;
    预处理单元,用于对获取到的舆情文本进行预处理,得到每个舆情文本的多个词组;
    确定单元,用于对每个舆情文本的多个词组进行命名实体识别,并将识别到企业实体的舆情文本确定为目标文本;
    分类单元,用于基于依存句法分析技术,对所述目标文本进行主体识别,得到识别结果,并按照预设的分类标准对所述目标文本进行分类,得到分类结果;
    输入单元,用于将所述目标文本、所述识别结果及所述分类结果输入至预先训练好的情感模型中,得到情感结果;
    选取单元,用于选取出情感结果为负面情绪的目标文本,作为预警文本;
    所述确定单元,还用于将所述预警文本与配置列表中的预警词组进行匹配,并将匹配到的预警词组确定为所述预警文本的预警类别;
    生成单元,用于获取所述预警文本的目标企业,并根据所述预警类别及所述预警文本生成所述目标企业的预警信息。
PCT/CN2020/098236 2020-04-29 2020-06-24 企业舆情分析方法、装置、电子设备及介质 WO2021217843A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010354452.5 2020-04-29
CN202010354452.5A CN111695033B (zh) 2020-04-29 2020-04-29 企业舆情分析方法、装置、电子设备及介质

Publications (1)

Publication Number Publication Date
WO2021217843A1 true WO2021217843A1 (zh) 2021-11-04

Family

ID=72476811

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/098236 WO2021217843A1 (zh) 2020-04-29 2020-06-24 企业舆情分析方法、装置、电子设备及介质

Country Status (2)

Country Link
CN (1) CN111695033B (zh)
WO (1) WO2021217843A1 (zh)

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114282542A (zh) * 2021-12-28 2022-04-05 中国农业银行股份有限公司 网络舆情监控方法及设备
CN114611515A (zh) * 2022-01-28 2022-06-10 江苏省联合征信有限公司 一种基于企业舆情信息识别企业实际控制人的方法和系统
CN114757191A (zh) * 2022-03-29 2022-07-15 国网江苏省电力有限公司营销服务中心 基于深度学习的电力舆情领域命名实体识别方法系统
CN114897038A (zh) * 2022-02-22 2022-08-12 平安科技(深圳)有限公司 目标词语的检测方法、装置、电子设备及存储介质
CN115759104A (zh) * 2023-01-09 2023-03-07 山东大学 基于实体识别的金融领域舆情分析方法与系统
CN115858738A (zh) * 2023-02-27 2023-03-28 浙江浙商金控有限公司 一种企业舆情信息相似性识别方法
CN115859773A (zh) * 2022-11-07 2023-03-28 中电科大数据研究院有限公司 一种基于群体情绪预测模型的预警方法、装置及存储介质
CN116013027A (zh) * 2022-08-05 2023-04-25 航天神舟智慧系统技术有限公司 一种群体性事件预警方法与系统
CN116342153A (zh) * 2023-05-31 2023-06-27 北京拓普丰联信息科技股份有限公司 一种提示的方法、装置、设备和介质
CN116522013A (zh) * 2023-06-29 2023-08-01 乐麦信息技术(杭州)有限公司 基于社交网络平台的舆情分析方法及系统
CN117009524A (zh) * 2023-08-08 2023-11-07 宇哲融创科技(北京)有限公司 一种基于舆情情感分析的互联网大数据分析方法及系统
CN117390184A (zh) * 2023-10-08 2024-01-12 南京特尔顿信息科技有限公司 一种基于大数据技术的网络舆情预警方法及系统
CN117407527A (zh) * 2023-10-19 2024-01-16 重庆邮电大学 一种教育领域舆情大数据分类方法
CN117453863A (zh) * 2023-12-22 2024-01-26 珠海博维网络信息有限公司 一种舆情文本归类方法及系统
CN117764458A (zh) * 2024-01-15 2024-03-26 广州宇中网络科技有限公司 业绩趋势分析方法、装置、设备及存储介质

Families Citing this family (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112183093A (zh) * 2020-11-02 2021-01-05 杭州安恒信息安全技术有限公司 一种企业舆情分析方法、装置、设备及可读存储介质
CN113297283A (zh) * 2020-11-12 2021-08-24 苏宁金融科技(南京)有限公司 用于企业风险预警的舆情分析方法及系统
CN112463963A (zh) * 2020-11-30 2021-03-09 深圳前海微众银行股份有限公司 识别目标舆情的方法、模型训练方法及装置
CN112434163A (zh) * 2020-11-30 2021-03-02 北京沃东天骏信息技术有限公司 风险识别方法及模型构建方法、装置、电子设备和介质
CN112541361A (zh) * 2020-12-11 2021-03-23 中信银行股份有限公司 一种针对特定业务对象的舆情处理方法及装置
CN112711651A (zh) * 2020-12-30 2021-04-27 上海金仕达软件科技有限公司 一种舆情监测方法及系统
CN112883734B (zh) * 2021-01-15 2023-01-10 成都链安科技有限公司 区块链安全事件舆情监测方法及系统
CN112860841B (zh) * 2021-01-21 2023-10-24 平安科技(深圳)有限公司 一种文本情感分析方法、装置、设备及存储介质
CN112905790A (zh) * 2021-02-04 2021-06-04 中国建设银行股份有限公司 监管事件定性指标提取的方法、装置和系统
CN113035240B (zh) * 2021-02-25 2024-02-02 平安普惠企业管理有限公司 语音播报方法、装置、设备及存储介质
CN113689299B (zh) * 2021-05-10 2023-10-20 深圳价值在线信息科技股份有限公司 新闻信息指数模型的构建方法及新闻信息分析方法
CN115512529B (zh) * 2021-06-23 2024-03-05 中国石油化工股份有限公司 承包商问题预警方法、预警装置及预警系统
CN113205817B (zh) * 2021-07-06 2021-12-07 明品云(北京)数据科技有限公司 语音语义识别方法、系统、设备及介质
CN113642867A (zh) * 2021-07-30 2021-11-12 南京星云数字技术有限公司 评估风险的方法及系统
CN113724830B (zh) * 2021-08-31 2024-04-30 深圳平安智慧医健科技有限公司 基于人工智能的用药风险检测方法及相关设备
CN113779251B (zh) * 2021-09-08 2024-04-19 平安国际智慧城市科技股份有限公司 线索信息获取方法、装置、介质及电子设备
CN113626718A (zh) * 2021-09-18 2021-11-09 广东电网有限责任公司广州供电局 一种企业管理系统人机交互事件处理方法及系统
CN113888760B (zh) * 2021-09-29 2024-04-23 平安银行股份有限公司 基于软件应用的违规信息监控方法、装置、设备及介质
CN114003694A (zh) * 2021-11-01 2022-02-01 北京明略软件系统有限公司 实体情感分析方法、系统、存储介质及电子设备
CN114386422B (zh) * 2022-01-14 2023-09-15 淮安市创新创业科技服务中心 基于企业污染舆情抽取的智能辅助决策方法及装置
CN116304058B (zh) * 2023-04-27 2023-08-08 云账户技术(天津)有限公司 企业负面信息的识别方法、装置、电子设备及存储介质
CN116738070A (zh) * 2023-08-15 2023-09-12 浙江同信企业征信服务有限公司 一种舆情监测方法、装置、设备及存储介质
CN117393003B (zh) * 2023-12-13 2024-03-15 广州云趣信息科技有限公司 一种基于语义理解的情绪识别方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110295594A1 (en) * 2010-05-28 2011-12-01 International Business Machines Corporation System, method, and program for processing text using object coreference technology
CN109583738A (zh) * 2018-11-22 2019-04-05 第创业证券股份有限公司 一种用于债券风险控制的装置及方法
CN110704622A (zh) * 2019-09-27 2020-01-17 北京明略软件系统有限公司 文本情感分类方法、装置及电子设备
CN110929026A (zh) * 2018-09-19 2020-03-27 阿里巴巴集团控股有限公司 一种异常文本识别方法、装置、计算设备及介质
CN111061876A (zh) * 2019-12-10 2020-04-24 中国建设银行股份有限公司 事件舆情数据分析方法及装置

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8417713B1 (en) * 2007-12-05 2013-04-09 Google Inc. Sentiment detection as a ranking signal for reviewable entities
CN106991090B (zh) * 2016-01-20 2020-12-11 北京国双科技有限公司 舆情事件实体的分析方法及装置
US20180108025A1 (en) * 2016-10-19 2018-04-19 Sap Se Systems and methods for sentiment insight in electronic media postings
CN108846547A (zh) * 2018-05-06 2018-11-20 成都信息工程大学 一种动态调整的企业信用风险评估方法
CN109582796A (zh) * 2018-12-05 2019-04-05 深圳前海微众银行股份有限公司 企业舆情事件网络的生成方法、装置、设备及存储介质
CN109684481A (zh) * 2019-01-04 2019-04-26 深圳壹账通智能科技有限公司 舆情分析方法、装置、计算机设备和存储介质
CN109992661A (zh) * 2019-03-05 2019-07-09 广发证券股份有限公司 一种面向证券行业的智能舆情监控方法及系统
CN111046172B (zh) * 2019-10-30 2024-04-12 北京奇艺世纪科技有限公司 一种舆情分析方法、装置、设备和存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110295594A1 (en) * 2010-05-28 2011-12-01 International Business Machines Corporation System, method, and program for processing text using object coreference technology
CN110929026A (zh) * 2018-09-19 2020-03-27 阿里巴巴集团控股有限公司 一种异常文本识别方法、装置、计算设备及介质
CN109583738A (zh) * 2018-11-22 2019-04-05 第创业证券股份有限公司 一种用于债券风险控制的装置及方法
CN110704622A (zh) * 2019-09-27 2020-01-17 北京明略软件系统有限公司 文本情感分类方法、装置及电子设备
CN111061876A (zh) * 2019-12-10 2020-04-24 中国建设银行股份有限公司 事件舆情数据分析方法及装置

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114282542A (zh) * 2021-12-28 2022-04-05 中国农业银行股份有限公司 网络舆情监控方法及设备
CN114611515A (zh) * 2022-01-28 2022-06-10 江苏省联合征信有限公司 一种基于企业舆情信息识别企业实际控制人的方法和系统
CN114611515B (zh) * 2022-01-28 2023-12-12 江苏省联合征信有限公司 一种基于企业舆情信息识别企业实际控制人的方法和系统
CN114897038A (zh) * 2022-02-22 2022-08-12 平安科技(深圳)有限公司 目标词语的检测方法、装置、电子设备及存储介质
CN114757191A (zh) * 2022-03-29 2022-07-15 国网江苏省电力有限公司营销服务中心 基于深度学习的电力舆情领域命名实体识别方法系统
CN116013027A (zh) * 2022-08-05 2023-04-25 航天神舟智慧系统技术有限公司 一种群体性事件预警方法与系统
CN115859773A (zh) * 2022-11-07 2023-03-28 中电科大数据研究院有限公司 一种基于群体情绪预测模型的预警方法、装置及存储介质
CN115759104B (zh) * 2023-01-09 2023-09-22 山东大学 基于实体识别的金融领域舆情分析方法与系统
CN115759104A (zh) * 2023-01-09 2023-03-07 山东大学 基于实体识别的金融领域舆情分析方法与系统
CN115858738A (zh) * 2023-02-27 2023-03-28 浙江浙商金控有限公司 一种企业舆情信息相似性识别方法
CN116342153A (zh) * 2023-05-31 2023-06-27 北京拓普丰联信息科技股份有限公司 一种提示的方法、装置、设备和介质
CN116522013B (zh) * 2023-06-29 2023-09-05 乐麦信息技术(杭州)有限公司 基于社交网络平台的舆情分析方法及系统
CN116522013A (zh) * 2023-06-29 2023-08-01 乐麦信息技术(杭州)有限公司 基于社交网络平台的舆情分析方法及系统
CN117009524A (zh) * 2023-08-08 2023-11-07 宇哲融创科技(北京)有限公司 一种基于舆情情感分析的互联网大数据分析方法及系统
CN117009524B (zh) * 2023-08-08 2024-03-05 宇哲融创科技(北京)有限公司 一种基于舆情情感分析的互联网大数据分析方法及系统
CN117390184A (zh) * 2023-10-08 2024-01-12 南京特尔顿信息科技有限公司 一种基于大数据技术的网络舆情预警方法及系统
CN117407527A (zh) * 2023-10-19 2024-01-16 重庆邮电大学 一种教育领域舆情大数据分类方法
CN117453863A (zh) * 2023-12-22 2024-01-26 珠海博维网络信息有限公司 一种舆情文本归类方法及系统
CN117453863B (zh) * 2023-12-22 2024-03-29 珠海博维网络信息有限公司 一种舆情文本归类方法及系统
CN117764458A (zh) * 2024-01-15 2024-03-26 广州宇中网络科技有限公司 业绩趋势分析方法、装置、设备及存储介质

Also Published As

Publication number Publication date
CN111695033A (zh) 2020-09-22
CN111695033B (zh) 2023-06-27

Similar Documents

Publication Publication Date Title
WO2021217843A1 (zh) 企业舆情分析方法、装置、电子设备及介质
Rauh Validating a sentiment dictionary for German political language—a workbench note
US10936970B2 (en) Machine learning document processing
Yu et al. Attention-based convolutional approach for misinformation identification from massive and noisy microblog posts
US8972408B1 (en) Methods, systems, and articles of manufacture for addressing popular topics in a social sphere
US11113323B2 (en) Answer selection using a compare-aggregate model with language model and condensed similarity information from latent clustering
US11762990B2 (en) Unstructured text classification
CN104471568A (zh) 对自然语言问题的基于学习的处理
CN112686022A (zh) 违规语料的检测方法、装置、计算机设备及存储介质
CN110929025A (zh) 垃圾文本的识别方法、装置、计算设备及可读存储介质
Salim et al. Deep LSTM-RNN with word embedding for sarcasm detection on Twitter
WO2022257452A1 (zh) 表情回复方法、装置、设备及存储介质
CN112805715A (zh) 识别实体属性关系
CN111522919A (zh) 一种文本处理方法、电子设备和存储介质
CN113064973A (zh) 文本分类方法、装置、设备及存储介质
CN112966117A (zh) 实体链接方法
CN110610003B (zh) 用于辅助文本标注的方法和系统
WO2021012958A1 (zh) 原创文本甄别方法、装置、设备与计算机可读存储介质
CN117351336A (zh) 图像审核方法和相关设备
Mounika et al. Design of book recommendation system using sentiment analysis
Chunamari et al. Quora question pairs using XG boost
Gong et al. Improving extractive document summarization with sentence centrality
Mahalakshmi et al. Twitter sentiment analysis using conditional generative adversarial network
CN113836941B (zh) 一种合同导航方法及装置
Halgamuge et al. Computation time optimization on hashtag segmentation for social media data

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: 20933223

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: 20933223

Country of ref document: EP

Kind code of ref document: A1