Enterprise public opinion monitoring processing method and device, computer equipment and storage medium

Abstract

The invention relates to an enterprise public opinion monitoring processing method and device based on archived big data, computer equipment and a storage medium. The method comprises the steps of obtaining to-be-processed information related to a target enterprise; preprocessing the to-be-processed information to obtain at least one group of topic events and public opinion data and credit data corresponding to the topic events, and determining hot events in the topic events; performing public opinion emotion prediction on the public opinion data and credit value prediction on the credit data according to the hot events, respectively obtaining corresponding prediction results, fusing the prediction results by utilizing an attention mechanism, and calculating to obtain an attention value between a public opinion emotion tendency value and a credit value prediction value; and according to the attention value, performing big data matching from a preset public opinion archive library to extract a corresponding guide scheme. The topic event and the public opinion development situation of the target enterprise company are found from the network news, and the trend of the network public opinion can be timely and correctly mastered, so that the public opinion development of the event is controlled, or the investment strategy is timely adjusted according to the public opinion development.

Description

Technical Field

[0001] The present application relates to the field of big data processing technology, and in particular to a method, apparatus, computer equipment and storage medium for monitoring and processing enterprise public opinion. Background Art

[0002] With the rapid development of the internet and new media, news and information are exploding exponentially. The volume of information released by news websites, industry forums, Weibo, and WeChat has reached an astonishing level. Consequently, any event, once reported and disseminated, has the potential to become a hot topic. This, through intentional or unintentional manipulation, can lead individuals to develop biased perceptions of the news. These biases can lead to public sentiment. When an event is linked to a target company, public opinion significantly influences its fluctuations, as creditworthiness is a barometer of a company's operating risk. If a target company's creditworthiness remains below its intrinsic value for a prolonged period, this can impact its ability to remarket and raise capital, and negatively impact the company's healthy operations.

[0003] Therefore, predicting hot news related to target companies, predicting their impact on credit scores based on public opinion data, and developing appropriate action plans based on these predictions is an extremely complex process. Traditional public opinion monitoring solutions typically employ semi-manual methods, such as using TF-IDF technology for news mining. However, due to the sparse feature information obtained, the large dimensionality of the data source, and low computational efficiency, they cannot effectively meet the requirements of timely detection. Therefore, traditional methods are difficult to meet the needs of target companies in the current internet era to handle capital market public opinion. Summary of the Invention

[0004] Based on this, the present application provides a corporate public opinion monitoring and processing method, device, computer equipment and storage medium based on archived big data, which can discover topic events and public opinion development trends of target corporate companies from online news, and can timely and correctly grasp the direction of online public opinion, so as to implement corresponding processing plans according to the development of public opinion, or adjust investment strategies in time according to the development of public opinion.

[0005] A method for monitoring and processing enterprise public opinion based on archived big data, the method comprising:

[0006] Obtain pending information related to the target enterprise;

[0007] Preprocessing the information to be processed to obtain at least one set of topic events, public opinion data corresponding to the topic events, and credit data;

[0008] Use clustering statistical algorithms to identify hot events in topic events;

[0009] For hot events, use the preset prediction model to predict public opinion sentiment on public opinion data to obtain the first prediction result, and predict the credit value of credit data to obtain the second prediction result;

[0010] The first prediction result and the second prediction result are integrated using the attention mechanism to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value;

[0011] According to the attention value between the public opinion sentiment tendency value and the credit value prediction value, big data matching is performed from the preset public opinion archive to extract the corresponding guidance plan.

[0012] In one embodiment, before the step of obtaining information to be processed related to a target enterprise, the following steps are included:

[0013] Use the visual interactive interface to customize the data publishing source;

[0014] Use crawler technology to crawl news information released by the target enterprise within a predetermined time period from data publishing sources, as well as relevant credit score information of the target enterprise. News information includes news headline content, news body content, netizen comments, netizen recognition index of comments, and dissemination popularity. Credit score information includes the score of the previous month, the score of the current month, the highest score, the lowest score, revenue, and order return rate.

[0015] Stored in the database according to the predetermined storage format.

[0016] In one embodiment, the step of pre-processing the news information includes:

[0017] Filter each piece of news information in the database according to the preset filtering rules;

[0018] According to the named entity rule, first feature information is extracted from the news title and the main content of the news, and a first joint feature set is formed;

[0019] Extracting the event item type with the highest similarity to the first joint feature set from the preset event library to represent each topic event;

[0020] And use the credit sentiment dictionary to extract the second feature information including credit sentiment words in the netizen comment content, and combine the second feature information with the netizen's recognition index and dissemination heat of the comment content to form a second joint feature set as public opinion data.

[0021] In one embodiment, the step of determining hot events in topic events using a clustering statistical algorithm includes:

[0022] According to the preset weight coefficients of each type of event item in the event library, after determining the topic event, the weight coefficient of the topic event is obtained;

[0023] Cluster events on the same topic and count the popularity of each topic event;

[0024] Determine the hot events among the topic events based on their weight coefficients and dissemination heat.

[0025] In one embodiment, the step of pre-processing the information to be processed further includes:

[0026] For hot events, the SimBERT algorithm is used to calculate the similarity of the second feature information in the second joint feature set to obtain a similarity matrix based on the number of public opinions. DBSCAN cluster analysis is performed on the similarity matrix to calculate netizens' recognition index and dissemination heat of the comments, and the second feature information with the largest proportion is obtained as the input data for the public opinion data model.

[0027] And use the normalization algorithm to normalize the credit data of the same period to obtain model input data of the credit data.

[0028] In one embodiment, the steps of using a preset prediction model to predict public opinion sentiment on public opinion data and predict credit value on credit data, and using an attention mechanism to fuse the prediction results to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value include:

[0029] Use the RoBERTa model to predict and derive the sentiment value of public opinion data;

[0030] Use the GRU model to predict the credit value at the predetermined time;

[0031] The query value is used to represent the sentiment tendency of public opinion, and the key-value value is used to represent the credit value prediction value. The correlation between each query value and each key value is calculated to obtain the weight coefficient of each key value corresponding to the value value; the Softmax function is used to normalize the weight coefficient; the weight and key value are weighted and summed to obtain the final attention value.

[0032] In one embodiment, based on the attention value between the public opinion sentiment tendency value and the credit value prediction value, a step of performing big data matching from a preset public opinion archive to extract a corresponding guidance solution includes:

[0033] Obtain the attention value between the public opinion sentiment tendency value and the credit value prediction value;

[0034] According to the preset public opinion archive, several guidance schemes are configured, and each guidance scheme is set with an attention value, so that the corresponding guidance scheme in the scheme library is called according to the attention value matching.

[0035] A device for monitoring and processing enterprise public opinion based on archived big data, comprising:

[0036] Information acquisition module, used to obtain the relevant information to be processed of the target enterprise;

[0037] A preprocessing module is used to preprocess the information to be processed to obtain at least one group of topic events, public opinion data corresponding to the topic events, and credit data;

[0038] A hotspot determination module is used to determine hot events in topic events using a clustering statistical algorithm;

[0039] The model prediction module is used to predict public opinion sentiment of public opinion data using a preset prediction model for hot events to obtain a first prediction result, and to predict credit value of credit data to obtain a second prediction result;

[0040] A prediction fusion module is used to fuse the first prediction result with the second prediction result using the attention mechanism to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value;

[0041] The solution extraction module is used to perform big data matching from the preset public opinion archive based on the attention value between the public opinion sentiment tendency value and the credit value prediction value to extract the corresponding guidance solution.

[0042] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are performed:

[0043] Obtain pending information related to the target enterprise;

[0044] Preprocessing the information to be processed to obtain at least one set of topic events, public opinion data corresponding to the topic events, and credit data;

[0045] Use clustering statistical algorithms to identify hot events in topic events;

[0046] For hot events, use the prediction model to predict public opinion sentiment on public opinion data to obtain the first prediction result, and predict the credit value of credit data to obtain the second prediction result;

[0047] The first prediction result and the second prediction result are integrated using the attention mechanism to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value;

[0048] According to the attention value between the public opinion sentiment tendency value and the credit value prediction value, big data matching is performed from the preset public opinion archive to extract the corresponding guidance plan.

[0049] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0050] Obtain pending information related to the target enterprise;

[0051] Preprocessing the information to be processed to obtain at least one set of topic events, public opinion data corresponding to the topic events, and credit data;

[0052] Use clustering statistical algorithms to identify hot events in topic events;

[0053] For hot events, use the prediction model to predict public opinion sentiment on public opinion data to obtain the first prediction result, and predict the credit value of credit data to obtain the second prediction result;

[0054] The first prediction result and the second prediction result are integrated using the attention mechanism to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value;

[0055] According to the attention value between the public opinion sentiment tendency value and the credit value prediction value, big data matching is performed from the preset public opinion archive to extract the corresponding guidance plan.

[0056] The above-mentioned enterprise public opinion monitoring and processing method, device, computer equipment and storage medium based on archived big data can include the following technical effects.

[0057] By monitoring and processing information related to a specific target company, it is possible to monitor the target company itself or other target companies of interest. This allows for early access to positive and negative data on public opinion trends regarding the target company of interest, enabling managers to promptly guide and control public opinion, and investors to promptly adjust future investment strategies, effectively reducing financial risks and taking effective preventive measures and remedial measures before a crisis occurs to avoid unnecessary losses. Investors can also use public opinion forecasts to evaluate target companies, clearly and intuitively understanding their potential value and whether they should change their investment direction in the future.

[0058] The system uses four stages: the first stage is data acquisition, which involves capturing credit score information and news information related to the target enterprise from at least one data source; the second stage is pre-processing of news information and credit score information to determine hot events, and the public opinion data corresponding to the hot events and the credit data of the same period are used as the input data of the model; the third stage uses the prediction model to obtain input data, and uses the relevant parameters of the enterprise's operating credit data as labels, uses the RoBERTa model to predict public opinion sentiment on the public opinion data, and the GRU model to predict credit data, and fuses the prediction results with the attention mechanism to obtain the attention value; the fourth stage extracts the corresponding guidance plan from the solution library based on the attention value. By using different prediction models to predict different data, the accuracy and precision can be improved, and the company's operating system can be more perfected. For target companies or investors, a complete risk warning system can be established to reduce the company's financial crisis and ensure the company's stable development.

[0059] This application has the advantages of simplicity, efficiency, and timeliness. It uses the hot events in online news and their development trends, and based on the advantages of archived big data storage, it sorts out various public opinion response plan data in advance to establish a public opinion archive. It can help relevant companies to grasp the direction of online public opinion in a timely and correct manner, control the development of events or guide public opinion in a timely manner, improve the stability of the business environment and maintain the image of relevant companies. It is of great significance and effectively improves user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 The figure is a flowchart of a method for monitoring and processing enterprise public opinion based on archived big data in one embodiment.

[0061] Figure 2 The figure is a simplified flowchart of a method for enterprise public opinion monitoring and processing based on archived big data in one embodiment.

[0062] Figure 3 The figure is a detailed flowchart of a method for monitoring and processing enterprise public opinion based on archived big data in one embodiment.

[0063] Figure 4 Schematic diagram of the input and output flow of the GRU model in one embodiment.

[0064] Figure 5 Schematic diagram of the internal structure of a GRU model in one embodiment.

[0065] Figure 6 Schematic diagram of the flow of the attention mechanism fusion method in one embodiment.

[0066] Figure 7 This is a structural block diagram of an enterprise public opinion monitoring and processing device based on archived big data in one embodiment. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0068] As businesses continue to grow, they generate massive amounts of data in their daily operations. This data includes, but is not limited to, customer information, transaction records, financial statements, employee profiles, and R&D materials. This data is fundamental to corporate decision-making, market analysis, internal management, and compliance audits. Without a systematic data archive to store and manage this data, businesses may face problems such as data loss, information leakage, and inefficiencies. Establishing an enterprise data archive can help companies centrally store and manage data, improving its security, reliability, and usability. Through the archive, companies can better organize, categorize, and index data for rapid retrieval and query. Furthermore, the archive can assist companies in data analysis and mining, providing data support for strategic decision-making.

[0069] Reference Attachment Figure 1-3 As shown, the enterprise public opinion monitoring and processing method based on archived big data provided by this application can be applied to the business credit risk monitoring of target enterprises. The enterprise public opinion monitoring and processing method based on archived big data includes the following steps.

[0070] Step S100, obtaining information to be processed related to the target enterprise;

[0071] Step S200: pre-processing the information to be processed to obtain at least one group of topic events, public opinion data corresponding to the topic events, and credit data;

[0072] Step S300, using a clustering statistical algorithm to determine hot events in topic events;

[0073] Step S400: For a hot event, use the prediction model to perform public opinion sentiment prediction on the public opinion data to obtain a first prediction result, and perform credit value prediction on the credit data to obtain a second prediction result;

[0074] Step S500: Using the attention mechanism, the first prediction result and the second prediction result are integrated to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value;

[0075] Step S600: Based on the attention value between the public opinion sentiment tendency value and the credit value prediction value, big data matching is performed from the preset public opinion archive to extract the corresponding guidance plan.

[0076] In step S100, which involves obtaining information to be processed related to a target enterprise, the target enterprise represents a target enterprise, which can be a company to be monitored, such as "Orient Securities," whose stock code is "600958." The target enterprise can be a listed company itself or a competitor. In this embodiment, the information to be processed includes at least news information and credit score information. This credit score information includes, but is not limited to, the previous month's score, the current month's score, the highest score, the lowest score, revenue, and return rate, among other parameters related to the enterprise's operational credit.

[0077] This step S100, before the step of obtaining the information to be processed related to the target enterprise, includes:

[0078] Use the visual interactive interface to customize the data release source; use crawler technology to crawl the news information released by the target enterprise within the predetermined time from the data release source, and crawl the credit score information related to the target enterprise, wherein the news information includes the news title content, the main content of the news, the content of netizens' comments, the netizens' recognition index of the comment content, and the degree of dissemination; the credit score information includes the score of the previous month, the score of the current month, the highest score, the lowest score, the revenue amount and the return rate; store it in the database according to the predetermined storage format. If the data release source in this step covers the topic events in the news information, the corresponding public opinion data and the credit data in the credit score information, then it can be customized to at least one data release source. The data release source in this step can be, but is not limited to, stock bars, Weibo, WeChat, Toutiao, Zhihu, news clients, etc.

[0079] In step S200, the steps of pre-processing the information to be processed include: filtering each piece of news information in the database according to preset filtering rules; extracting first feature information from the news title and the main content of the news according to the named entity rule to form a first joint feature set; extracting the event item type with the highest similarity to the first joint feature set from the preset event library to represent each topic event; and using the credit sentiment dictionary to extract second feature information including credit sentiment words from the netizen comments, and combining the second feature information with the netizen's recognition index of the comment content and the degree of dissemination to form a second joint feature set as public opinion data. The event items pre-stored in the event library include, but are not limited to, personnel risk, liquidity risk, performance risk, credit risk, and product risk.

[0080] Step S300: using a clustering statistical algorithm to determine hot events in topic events.

[0081] Among them, the steps of determining hot events in topic events using a clustering statistical algorithm include: obtaining the weight coefficient of a topic event after determining the topic event based on the preset weight coefficients of each type of event item in the event library; performing clustering processing on the same topic event and counting the spread heat of each topic event; and determining the hot events in the topic events based on the weight coefficient and spread heat of the topic event.

[0082] In step S200, the step of pre-processing the information to be processed further includes:

[0083] For hot events, the SimBERT algorithm is used to calculate the similarity of the second feature information in the second joint feature set to obtain a similarity matrix based on the number of public opinions; DBSCAN cluster analysis is performed on the similarity matrix to count netizens' recognition index and dissemination heat of the comment content, and obtain the second feature information with the largest proportion as the model input data of the public opinion data; and the normalization algorithm is used to normalize the credit data to obtain the model input data of the credit data.

[0084] Among them, using the SimBERT algorithm to calculate the similarity of the second feature information in the second joint feature set to obtain a similarity matrix based on the number of comments, the main application scenarios of the SimBERT algorithm include the generation of similar texts and the retrieval of similar texts, so similar sentences can be generated or sentence similarity can be determined. In this embodiment, the SimBERT algorithm is used to perform a dot product calculation on the sentence vectors generated by BertPredic in the second feature information, thereby calculating the similarity between the second feature information and obtaining an N*N similarity matrix, where N is the number of comments.

[0085] The second feature information represents text information containing credit sentiment words from the credit sentiment dictionary. In other words, this embodiment uses the credit sentiment dictionary to extract text information containing credit sentiment words from netizen comments. This second feature information is then combined with the netizen's approval index and virality of the comments to form a second joint feature set to represent public opinion data.

[0086] The characteristic words in the credit sentiment dictionary in this embodiment may include: praise, criticism, plus points, minus points, integrity, responsibility, liar, routine, etc., which can be further divided into positive, neutral, and negative. By obtaining different characteristic words, the emotional tendency corresponding to the second characteristic information can be obtained.

[0087] In this step, DBSCAN cluster analysis is performed on the similarity matrix to obtain the second feature information with the largest proportion, which is used as the model input data of the public opinion data.

[0088] DBSCAN cluster analysis refers to the use of the DBSCAN algorithm to perform a maximum density-connected sample set derived from a density-reachable relationship to represent a cluster of the final cluster. The execution steps of the DBSCAN algorithm include: inputting a second feature information data set consisting of a similarity matrix, setting the neighborhood radius Eps, and the threshold value of the number of data objects in the neighborhood MinPts; arbitrarily selecting a data object point p from the second feature information data set; for the parameters Eps and MinPts, if the selected data object point p is a core point, then find all data object points that are density-reachable from p to form a cluster; if the selected data object point p is an edge point, then select another data object point; repeat the selection process until all points are processed; and output a density-connected cluster.

[0089] In this embodiment, DBSCAN clustering analysis is performed based on the similarity matrix generated by SimBERT to find the largest cluster, and the second feature information corresponding to the largest cluster is used as the model input data of the public opinion data.

[0090] The step of normalizing the credit data using a normalization algorithm to obtain model input data for the credit data includes:

[0091] Based on the different dimensions of transaction prices and transaction volumes of orders in the previous month in the credit data, the credit data in the database is normalized. Specifically,

[0092] All data x for each indicator x in the credit data i , select the maximum value x max and the minimum value x min , transform each credit data so that all credit data fall into the interval [0, 1]. The formula is as follows:

[0093]

[0094] Among them, x represents the credit data before normalization, x * represents the normalized credit data, x min Represents the minimum value in the original credit data, x max Indicates the maximum value in the original data. In order to compare the prediction results, it is necessary to perform denormalization. The formula is as follows:

[0095] y=y * (y max -y min )+y min

[0096] Where y represents the predicted value of the prediction model after denormalization, y * Represents the predicted value of the prediction model, y minRepresents the minimum value of the original credit data, y max Indicates the maximum value of credit data.

[0097] Step S400: For hot events, use a preset prediction model to perform public opinion sentiment prediction on the public opinion data to obtain a first prediction result, and perform credit value prediction on the credit data to obtain a second prediction result.

[0098] The public opinion sentiment prediction results from public opinion data serve as the basis for determining the severity of public opinion surrounding a topic event. This embodiment employs a dependency-based sentiment recognition method to analyze public opinion data related to the same topic event and obtain fine-grained sentiment analysis results. Specifically, in addition to the content of netizen comments themselves, these analysis also includes influencing factors such as the netizen's approval index for the comments, the amount of dissemination, the amount of public opinion information, and the number of reposts.

[0099] In this embodiment, the steps of using a preset prediction model to predict public opinion sentiment for public opinion data and predict credit value for credit data, and using an attention mechanism to fuse the prediction results to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value include:

[0100] The RoBERTa model is used to predict and obtain the public opinion sentiment tendency value of the public opinion data; the Query value is used to represent the public opinion sentiment tendency value, and the Key-Value value is used to represent the credit value prediction value. Through the mapping function composed of multiple Query values and Key-Value values, the correlation between the Query value and each Key value is calculated to obtain the weight coefficient of the Value value corresponding to each Key value, and then the Softmax function is used to normalize the weight, and the processed weight is weighted and summed with the corresponding Key value to obtain the final attention value.

[0101] Among them, the step of using the RoBERTa model to predict and obtain the public opinion sentiment tendency value of the public opinion data includes using the RoBERTa model to predict the public opinion sentiment of the second feature information. Specifically,

[0102] The previous month's score, current month's score, highest score, lowest score, revenue, and return rate in the same time period are used as labels for the RoBERTa model. The text classification based on the RoBERTa model is fine-tuned to analyze the second feature information.

[0103] The RoBERTa model can achieve a high classification accuracy even when the credit information text sentiment analysis annotation dataset is very small. The RoBERTa model in this embodiment is a more effective pre-training model training method based on the BERT model, and has the same architecture as the BERT model. The RoBERTa model is an unsupervised pre-training language representation method that can well model the semantic information of text sentences and can solve the problem of the lack of large-scale and high-quality annotated datasets for credit information text. A large number of experiments have shown that under the same model structure as BER, the effect of the RoBERTa model on multiple NLP tasks can be further improved. This embodiment uses the RoBERTa model to extract the sentiment features of credit information text to obtain the public opinion sentiment tendency value of public opinion data.

[0104] Compared to the BERT model, the RoBERTa model is a bidirectional Transformer encoder based on the bidirectional Transformer model, using the encoder module in the Transformer for calculation. Each encoder used in the RoBERTa model in this embodiment has three parts: input, attention mechanism, and feedforward neural network (deep learning model).

[0105] Furthermore, when using the RoBERTa model to predict the sentiment value of public opinion data, the second feature information is converted into a token sequence and input into the deep learning model. The token can be represented as a token, which can be a Chinese character, word, or punctuation mark, and each token corresponds to a vector.

[0106] The deep learning model uses the positional relationship between the input vector and the corresponding token to calculate the polarity of the second feature information (text sentiment, or emotional feature). The calculated polarity is compared with the actual polarity to obtain the error. The error derivative is backpropagated to update the deep learning model parameters and token vectors. In this way, the calculated polarity of the given input continuously approaches the actual polarity.

[0107] Before the step of converting the second feature information into a token sequence in this embodiment, the step also includes performing data cleaning, filtering invalid information according to a preset data length, for example, removing stop words, and then using a tokenizer to segment the second feature information and convert it into a token sequence.

[0108] Use the GRU model to predict and obtain the credit value prediction value at the predetermined time.

[0109] Among them, refer to the attached Figure 4-5 As shown in the figure, in the step of using the GRU model to predict the credit value, GRU stands for gated recurrent unit, which is a recurrent neural network used to solve the gradient in long-term memory and back propagation. Figure 4 (a) shows the input and output structure of the GRU model, including a current input x t , the hidden state h passed down from the previous node t-1 Among them, the hidden state contains the relevant information of the previous node. Combined with x t and h t-1 , get the output y of the current hidden node through the GRU model t and the hidden state h passed to the next node t .

[0110] According to the last transmitted state h t-1 and the input x of the current node t , get two gate states. r represents the reset gate, and z represents the update gate. Figure 4 (b) The schematic diagram of the r and z gate control is shown. After obtaining the gate signal, the reset gate is used to obtain the data h after the "reset" t-1‘ =h t-1 ⊙r, then h t-1‘ With input x t Splice and then use a tanh activation function to scale the data to the range of -1 to 1, and you can get the following Figure 4 (c) shows h with input and output ′ .

[0111] Compared with LSTM, GRU has one less "gate" inside and fewer parameters than LSTM, but it can also achieve the same functions as LSTM.

[0112] Among them, (1-Z)⊙h t-1 : Indicates the selective "forgetting" of the original hidden state. You can think of (1-Z) as a forget gate, forget h t-1 Some unimportant information in the dimension. ′ Indicates the h containing the current node information ′ Perform selective "memory", 1-Z also means forget h ′ Some unimportant information in the dimension, thereby realizing h ′ Filtering of certain information in a dimension. t =(1-Z)⊙h t-1 +Z⊙h ′ Indicates that h was forgotten to be passed down t-1 and adds some dimensional information of the current node input.

[0113] like Figure 5The internal structure diagram of the GPU model shown in FIG. 1 shows that compared to the LSTM (Long Short Term Memory) network that uses multiple gates, the GPU model in this embodiment uses the same gate Z to simultaneously perform forgetting and selective memory, thereby simplifying computing power and reducing computing time.

[0114] When using the GRU model to predict the credit value at a predetermined time, the first step is to determine its input and output sequences. Unlike traditional neural networks that require data from each lag time from t-1 to tn as multiple feature inputs, the GRU network can directly use the entire time series as a single feature due to its unique memory structure, which greatly reduces the network structure and improves the calculation efficiency of node parameters. The input data x in this article mainly consists of the following features: score of the previous month, score of the current month, highest score, lowest score, revenue and return rate, the day of the month and the hour of the day, respectively. and express.

[0115] Since the data to be predicted has a great correlation with the data in the previous period, the characteristic data of each week in the previous period is selected as input, and the output is the data to be predicted.

[0116] After the input and output data are determined, they need to be normalized. This not only eliminates the impact of different dimensions between features but also allows the model to converge quickly. Each feature sequence is normalized to the range [0, 1]. Other features are represented by dummy variables, where 0 represents no and 1 represents yes.

[0117] The normalization formula is: Where: x t and are the values before and after normalization respectively; x max and x min The output of the previous hidden layer is used as the input of the next hidden layer to achieve deep extraction of the input. Finally, the output is normalized to obtain the credit value prediction values at four moments.

[0118] Step S500: Use the attention mechanism to fuse the first prediction result and the second prediction result to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value.

[0119] The step of using the attention mechanism to fuse the first prediction result and the second prediction result to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value further includes: using the Query value to represent the public opinion sentiment tendency value, using the Key-Value value to represent the credit value prediction value, calculating the correlation between each Query value and each Key value, and obtaining the weight coefficient of the Value value corresponding to each Key value; using the Softmax function to normalize the weight coefficient; and weighted summing the weight and the Key value to obtain the final attention value.

[0120] Specifically, if Figure 6 The fusion process diagram shown in the figure shows that this step uses the attention mechanism to fuse the public opinion sentiment tendency value and the credit value prediction value. The public opinion sentiment tendency value is set as the query (Query), and the credit value prediction value is set as the key-value pair (Key-Value). The specific fusion process includes the following steps:

[0121] Use formula (1) to calculate the correlation between each Query value and each Key value, and obtain the weight coefficient of each Key value corresponding to the Value value;

[0122]

[0123] Use the formula (2) of the Softmax function to normalize the weight coefficients;

[0124]

[0125] Use formula (3) to add the weight and key value to get the final attention value;

[0126]

[0127] Where A represents the final attention value, Lx is the data length, Q represents the query, K represents the key, and V represents the value.

[0128] Step S600, based on the attention value between the public opinion sentiment value and the credit value prediction value, performs big data matching from a preset public opinion archive to extract the corresponding guidance solution, including:

[0129] Obtain the attention value between the public opinion sentiment tendency value and the predicted value;

[0130] According to the preset public opinion archive, several guidance schemes are configured, and each guidance scheme is set with an attention value, so that the corresponding guidance scheme in the scheme library is called according to the attention value matching.

[0131] In the above-mentioned enterprise public opinion monitoring and processing method based on archived big data,

[0132] For the pending information related to a certain target enterprise, the public opinion theme is not limited. According to the specific object (for example, a specific target enterprise company), information retrieval is carried out from multiple data publishing sources to obtain relevant news information, so that hot topics can be obtained in a timely manner.

[0133] The information to be processed is pre-processed to obtain credit data, at least one group of topic events and their corresponding public opinion data, and a clustering statistical algorithm is used to determine the hot events in the topic events. Therefore, it can be known that the present application combines topic detection and public opinion sentiment recognition to monitor public opinion. Therefore, the present application divides corporate public opinion monitoring based on archived big data into two subtasks: hot topic detection and sentiment recognition of public opinion data, so that investors' emotional attitudes towards the target corporate companies can be discovered in a timely manner. And in this application, according to the named entity rule, the first feature information in the news title and the main content of the news is extracted to form a joint feature set; from the preset event library, the event item with the highest similarity to the joint feature set is extracted to represent each topic event; using the credit sentiment dictionary, the second feature information in the netizen comment content is extracted to represent the public opinion data. The accuracy of topic detection is improved by the fusion of multiple features for news representation;

[0134] For the hot events, a preset prediction model is used to predict public opinion sentiment and credit value for the public opinion data, and the attention mechanism is used to fuse the prediction results to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value. In other words, the correlation between the public opinion sentiment tendency value and the credit value fluctuation is calculated and determined. Based on the attention value between the public opinion sentiment tendency value and the credit value prediction value, the need to extract the corresponding guidance plan is determined. If it is an investor, a corresponding investment plan can be generated based on the attention value. If it is the target enterprise itself, a corresponding public opinion response guidance plan can be generated based on the attention value.

[0135] In one embodiment, Figure 7 As shown, a corporate public opinion monitoring and processing device based on archived big data is provided, comprising: an information acquisition module 100, a preprocessing module 200, a hotspot determination module 300, a model prediction module 400, a prediction fusion module 500, and a solution extraction module 600, wherein:

[0136] The information acquisition module 100 is used to acquire the to-be-processed information related to the target enterprise.

[0137] The preprocessing module 200 is used to preprocess the information to be processed to obtain at least one group of topic events, public opinion data corresponding to the topic events, and credit data.

[0138] The hotspot determination module 300 is used to determine hot events in topic events by using a clustering statistical algorithm.

[0139] The model prediction module 400 is used to use the prediction model to predict public opinion sentiment on public opinion data for hot events to obtain a first prediction result, and to predict the credit value of credit data to obtain a second prediction result.

[0140] The prediction fusion module 500 is used to use the attention mechanism to fuse the first prediction result and the second prediction result, and calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value.

[0141] The solution extraction module 600 is used to perform big data matching from a preset public opinion archive based on the attention value between the public opinion sentiment tendency value and the predicted value to extract the corresponding guidance solution.

[0142] Regarding the specific definition of the enterprise public opinion monitoring and processing device based on archived big data, please refer to the definition of the enterprise public opinion monitoring and processing method based on archived big data above, which will not be repeated here. The various modules in the above-mentioned enterprise public opinion monitoring and processing device based on archived big data can be implemented in whole or in part through software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0143] In one embodiment, a computer device is provided, which may be a server. The computer device includes a processor, a memory, a network interface, and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store xxx data. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for monitoring and processing enterprise public opinion based on archived big data is implemented.

[0144] In one embodiment, a computer device is provided, which may be a terminal. The computer device includes a processor, memory, a network interface, a display screen, and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for monitoring and processing enterprise public opinion based on archived big data is implemented. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen. The input device of the computer device may be a touch layer covering the display screen, or may be a key, trackball, or touchpad provided on the computer device housing, or may be an external keyboard, touchpad, or mouse.

[0145] Those skilled in the art will understand that the drawings provided in the present application do not constitute a limitation on the computer device to which the present application is applied. The specific computer device may include more or fewer components than shown in the drawings, or combine certain components, or have a different arrangement of components.

[0146] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0147] Obtain pending information related to the target enterprise;

[0148] Preprocessing the information to be processed to obtain at least one set of topic events, public opinion data corresponding to the topic events, and credit data;

[0149] Use clustering statistical algorithms to identify hot events in topic events;

[0150] For hot events, use the prediction model to predict public opinion sentiment on public opinion data to obtain the first prediction result, and predict the credit value of credit data to obtain the second prediction result;

[0151] The first prediction result and the second prediction result are integrated using the attention mechanism to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value;

[0152] According to the attention value between the public opinion sentiment tendency value and the predicted value, big data matching is performed from the preset public opinion archive to extract the corresponding guidance plan.

[0153] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0154] Obtain pending information related to the target enterprise;

[0155] Preprocessing the information to be processed to obtain at least one set of topic events, public opinion data corresponding to the topic events, and credit data;

[0156] Use clustering statistical algorithms to identify hot events in topic events;

[0157] For hot events, use the prediction model to predict public opinion sentiment on public opinion data to obtain the first prediction result, and predict the credit value of credit data to obtain the second prediction result;

[0158] The first prediction result and the second prediction result are integrated using the attention mechanism to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value;

[0159] According to the attention value between the public opinion sentiment tendency value and the credit value prediction value, big data matching is performed from the preset public opinion archive to extract the corresponding guidance plan.

[0160] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0161] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0162] The above embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for monitoring and processing corporate public opinion, characterized in that: The method comprises: Obtain pending information related to the target enterprise; Preprocessing the information to be processed to obtain at least one group of topic events, public opinion data corresponding to the topic events, and credit data; Using a clustering statistical algorithm to determine hot events in the topic events; For the hot event, using the prediction model to perform public opinion sentiment prediction on the public opinion data to obtain a first prediction result, and performing credit value prediction on the credit data to obtain a second prediction result; The first prediction result and the second prediction result are integrated using the attention mechanism to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value; According to the attention value between the public opinion sentiment tendency value and the credit value prediction value, big data matching is performed from the preset public opinion archive to extract the corresponding guidance plan.

2. The enterprise public opinion monitoring and processing method according to claim 1, characterized in that: The steps to obtain the information to be processed about the target enterprise include: Use the visual interactive interface to customize the data publishing source; Using crawler technology to crawl news information published by the target enterprise within a predetermined time from the data publishing source, as well as information related to the target enterprise, including credit score information, wherein the news information includes news headline content, news body content, netizen comments, netizen recognition index of the comments, and dissemination popularity; the credit score information includes the score of the previous month, the score of the current month, the highest score, the lowest score, revenue, and return rate; Stored in the database according to the predetermined storage format.

3. The enterprise public opinion monitoring and processing method according to claim 2, characterized in that: The steps of preprocessing news information include: Filtering each piece of news information in the database according to preset filtering rules; Extracting first feature information from the news title and the news body content according to named entity rules, and forming a first joint feature set; Extracting the event item type with the highest similarity to the first joint feature set from a preset event library to represent each topic event; And using the credit sentiment dictionary, extract the second feature information including the credit sentiment words in the netizen comment content, and combine the second feature information with the netizen's recognition index and dissemination heat of the comment content to form a second joint feature set as public opinion data.

4. The enterprise public opinion monitoring and processing method according to claim 3 is characterized in that: The step of determining the hot events in the topic events by using a clustering statistical algorithm includes: Determining the weight coefficient of the topic event according to the preset weight coefficients of each type of event item in the event library; Clustering is performed on the same topic event, and the spread popularity of each topic event is counted; According to the weight coefficient of the topic event and the dissemination heat, a hot event in the topic event is determined.

5. The enterprise public opinion monitoring and processing method according to claim 4, characterized in that: The step of preprocessing the information to be processed further includes: For the hot event, use the SimBERT algorithm to perform similarity calculation on the second feature information in the second joint feature set to obtain a similarity matrix based on the number of public opinions; Performing DBSCAN cluster analysis on the similarity matrix to calculate netizens' recognition index and dissemination popularity of the comments, obtaining the second feature information with the largest proportion, and using it as model input data for public opinion data; And using a normalization algorithm to normalize the credit data of the same period to obtain model input data of the credit data.

6. The method for monitoring and processing public opinion according to claim 1, characterized in that: The steps of using a prediction model to predict public opinion sentiment on the public opinion data and predicting credit value on the credit data, and using an attention mechanism to fuse the prediction results to calculate an attention value between the public opinion sentiment tendency value and the credit value prediction value include: Using the RoBERTa model to predict, the sentiment tendency value of the public opinion data is obtained; Use the GRU model to predict the credit value at the predetermined time; The Query value is used to represent the sentiment tendency value of public opinion, and the Key-Value value is used to represent the credit value prediction value. The correlation between each Query value and each Key value is calculated to obtain the weight coefficient of each Key value corresponding to the Value value; the Softmax function is used to normalize the weight coefficient; the weight and the Key value are weighted and summed to obtain the final attention value.

7. The method for monitoring and processing public opinion according to claim 1, characterized in that: According to the attention value between the public opinion sentiment tendency value and the predicted value, a step of performing big data matching from a preset public opinion archive to extract a corresponding guidance plan includes: Obtaining an attention value between the public opinion sentiment tendency value and the credit value prediction value; According to the preset public opinion archive, several guidance schemes are configured, and each guidance scheme is correspondingly set with an attention value, so that the corresponding guidance scheme in the scheme library is called according to the attention value matching.

8. An enterprise public opinion monitoring and processing device, characterized in that: The device comprises: Information acquisition module, used to obtain the relevant information to be processed of the target enterprise; A preprocessing module, configured to preprocess the information to be processed to obtain at least one group of topic events, public opinion data corresponding to the topic events, and credit data; A hotspot determination module, configured to determine hotspot events among the topic events using a clustering statistical algorithm; A model prediction module is used to use a prediction model to perform public opinion sentiment prediction on the public opinion data for the hot event to obtain a first prediction result, and to perform credit value prediction on the credit data to obtain a second prediction result; A prediction fusion module is used to fuse the first prediction result with the second prediction result using the attention mechanism to calculate the attention value between the public opinion sentiment tendency value and the credit value prediction value; The solution extraction module is used to perform big data matching from a preset public opinion archive based on the attention value between the public opinion sentiment tendency value and the credit value prediction value to extract the corresponding guidance solution.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images