Enterprise public opinion detection system

Abstract

A corporate public opinion detection system includes a processing unit. The processing unit categorizes multiple news items into at least one news category containing these news items. The processing unit analyzes the news items within each news category using a natural language model to generate multiple key phrases corresponding to each news item. The processing unit inputs multiple word and phrase data from a word database and the key phrases corresponding to the news items into a recognition model for analysis to generate multiple keywords corresponding to the news category within the news items. Based on these keywords, the processing unit generates multiple public opinion information items and compares the quantity of each public opinion information item with a threshold. When the processing unit determines that the quantity exceeds the threshold, it generates a risk list.

Description

[Technical Field]

[0001] This invention relates to a detection system, and more particularly to a corporate public opinion detection system. [Previous Technology]

[0002] Currently, when banks conduct risk monitoring of various enterprises and industries, they first examine relevant public opinion news through the enterprises and industries, and then the staff judges the degree of risk impact of these public opinion news on the enterprises and industries through expert opinions and experience.

[0003] However, by the time enterprises and industries review relevant public opinion news and then assess its risk status, the event has often already begun to ferment, making it difficult to prevent in advance. Moreover, the above process is quite demanding on the experience and judgment of the personnel handling it. Furthermore, the large amount of public opinion news content is also quite labor-intensive and may lead to human misjudgment or oversight, so there is a real need for improvement. [Summary of the Invention]

[0004] Therefore, the purpose of this invention is to provide an enterprise public opinion detection system that can automatically analyze recent public opinion news in an electronic manner to generate relevant keywords, and determine the relevant public opinion events and enterprises and industries that need to be monitored for risks based on the keywords, and generate a corresponding risk list. This can improve the efficiency of processing and enable banks to manage risks as early as possible based on relevant public opinion events. The standardized process can avoid unnecessary human errors in the process of manual comparison and modification, and at least solve the problem that the previous technology was not effective when processed manually.

[0005] Therefore, the enterprise public opinion detection system of the present invention is applicable to a data source server and includes a communication unit, a storage unit, and a processing unit. The communication unit is used to provide networking functionality to communicate with the data source server. The storage unit stores a natural language model, a recognition model, a word database, and a threshold. The processing unit is electrically connected to the communication unit and the storage unit. The processing unit downloads multiple news items from the data source server via the communication unit, classifies these news items into at least one news category containing them, and analyzes the news items in the news category using the natural language model to generate multiple important phrases corresponding to the news items. The processing unit reads the word database from the storage device, and inputs multiple phrases from the word database and the important phrases corresponding to the news items into the recognition model for analysis to generate multiple keywords in the news items corresponding to the news category. The processing unit generates multiple public opinion information items based on the keywords. Each public opinion information item contains a key event and a corresponding number of key events. The processing unit compares the number of each public opinion information item with a threshold. When the processing unit determines that the value of the number is greater than the threshold, the processing unit generates a risk list based on the value of the public opinion information item and the key event corresponding to the value.

[0006] In some embodiments, the storage unit also stores an exclusion list including multiple specific words, and the processing unit checks the news materials against the specific words. When the processing unit finds that any news material contains any of the specific words, the processing unit excludes the news material.

[0007] In some embodiments, the processing unit identifies each keyword as either a risk keyword or an industry keyword. The processing unit generates public opinion information based on either the risk keyword or the industry keyword. The processing unit combines the risk keyword and the industry keyword to generate the key event, and the processing unit compares the news data contained in the news category based on the risk keyword and the industry keyword. When the processing unit finds that any news data contains both the risk keyword and the industry keyword, the processing unit determines that the news data as target data, and the processing unit counts the target data to generate the number corresponding to the key event.

[0008] In some implementations, the processing unit uses the natural language model to determine the various parts of speech and frequencies corresponding to the important phrases.

[0009] In some embodiments, the news materials also include multiple release times. The processing unit determines a weight data corresponding to the news material based on the release time of each news material and a current time, and the processing unit adds the weight data to the corresponding news material.

[0010] In some embodiments, it is also applicable to an internal server of a bank, which is communicatively connected to the communication unit. After the processing unit generates the risk list, the processing unit transmits the risk list to the internal server via the communication unit.

[0011] In some embodiments, the processing unit downloads multiple industry data from the internal server via the communication unit. Each industry data includes an industry category and an industry name, and the processing unit checks the industry data against industry keywords. When the processing unit finds that either the industry category or the industry name of one of the industry data matches the industry keyword, the processing unit defines the industry data as a risky industry data and sends the risky industry data back to the internal server.

[0012] In some implementations, when the processing unit determines the weight data based on each piece of news material, the value of the weight data is inversely proportional to the difference between the publication time and the current time.

[0013] The advantages of this invention are as follows: Through the above process, the corporate public opinion detection system of this invention can download the news data at the fixed time, automatically classify the news data into news categories, generate important phrases according to the natural language model, and then generate public opinion data and risk list by analyzing the words and phrases and important phrases according to the recognition model and generating keywords, and send them back to the internal server for use by the processing personnel. It can improve the efficiency of processing, prevent the failure to handle events in time when they ferment, and reduce the risk of possible human misjudgment and oversight, thereby solving the problem that the previous technology was not effective when processed manually.

Implementation Method

[0015] Before the present invention is described in detail, it should be noted that, unless otherwise defined, the term "electrically connected" in this patent specification is used to describe the "coupled" relationship between computer hardware (e.g., electronic systems, devices, apparatuses, units, components), and generally refers to "wired electrical connections" achieved by physically connecting multiple computer hardware components through conductor / semiconductor materials, and "radio connections" that achieve wireless data transmission using wireless communication technologies (e.g., but not limited to wireless networks, Bluetooth, and electromagnetic induction). On the other hand, unless otherwise defined, the term "electrical connection" in this patent specification also generally refers to "direct electrical connections" achieved by directly coupling multiple computer hardware components to each other, and "indirect electrical connections" achieved by indirectly coupling multiple computer hardware components through other computer hardware components.

[0016] Before the present invention is described in detail, it should be noted that the term "unit" in this patent specification refers to computer hardware rather than software. For example, "processing unit" is used to represent computer hardware with data processing capabilities. On the other hand, the term "unit" in this patent specification can refer to a single computer hardware with a specific function, or it can refer to a group of computer hardware with similar functions. For example, "processing unit" can refer to a single processor with data processing capabilities, but it can also refer to a collection of processors.

[0017] Referring to Figure 1, one embodiment of the enterprise public opinion detection system 1 of the present invention is applicable to a data source server 8 and an internal server 9 of a bank, and includes a communication unit 2, a storage unit 3, and a processing unit 4. The data source server 8 is, for example, a news website (e.g., but not limited to, the BBC News Chinese website) and a social media platform (e.g., but not limited to, a news page on the Facebook platform), and is used to provide multiple news items, each containing multiple publication times.

[0018] The communication unit 2 is, for example, a chip or device that supports wired network technology (such as Ethernet) or wireless network technology (such as Wi-Fi) and is used to provide networking functionality so that it can communicate with the computer device (not shown) of the data source server 8 and the computer device (not shown) of the internal server 9 through the network, for example, by transmitting or receiving data with the computer device of the data source server 8 and the computer device of the internal server 9 through a local area network, an intranet, or the Internet.

[0019] The storage unit 3 is, for example, a memory, a hard disk, or other storage device, and stores a natural language model, a recognition model, a word database including multiple words and phrases, a threshold, and an exclusion list including multiple specific words.

[0020] The processing unit 4 is, for example, a central processing unit or a microprocessor, and is electrically connected to the communication unit 2 and the storage unit 3. At a predetermined fixed time (e.g., but not limited to 7:00 AM every day), the processing unit 4 communicates with the data source server 8 through the communication unit 2, and the processing unit 4 downloads the news data from the data source server 8.

[0021] Next, the processing unit 4 reads the exclusion list from the storage unit 3, and the processing unit 4 checks the news materials against the specific words to filter out each news material that covers the specific words (e.g., but not limited to elections or games). When the processing unit 4 finds that any news material contains any specific word, the processing unit 4 excludes the news material.

[0022] In addition, the processing unit 4 uses a BIRCH (Balanced Iterative Reducing and Clustering using Hierarchies) algorithm to generate multiple preset phrases for each news data, and classifies the news data into at least one news category containing the news data (e.g., but not limited to real estate risk category or chip bill category) by using the preset phrases, so as to automatically classify similar news data into the same news category, thereby avoiding the situation where the content of the news data is very different.

[0023] Furthermore, the processing unit 4 reads the natural language model from the storage unit 3 to analyze the news data in the news category using the natural language model. During the analysis, the processing unit 4 uses a TTF-IDF (Temporal Term Frequency – Inverse Document Frequency) technique to determine a weight data corresponding to each news data based on the publication time and the current time. The value of the weight data is inversely proportional to the difference between the publication time and the current time. For example, the greater the distance between the publication time and the current time, the smaller the value of the weight data; conversely, the closer the distance between the publication time and the current time, the larger the value of the weight data. The processing unit 4 also adds the weight data to the corresponding news data. Furthermore, the processing unit 4 uses the TTF-IDF technique to determine multiple frequencies corresponding to the preset word groups. The processing unit 4 uses KeyBERT (Key Bidirectional Encoder Representations from Transformers) technology to determine the part of speech (e.g., subject, verb, or object) of the preset phrases. The processing unit 4 also uses NER (Name Entity Recognition) prediction technology to determine whether the preset words have specific meanings (e.g., but not limited to names, place names, dates, and times). Finally, after processing with TTF-IDF, KeyBERT, and NER prediction technologies, the natural language model extracts and generates multiple important phrases from the preset phrases, each corresponding to a specific news item and containing the relevant surname and frequency.

[0024] Next, the processing unit 4 reads the word database and the recognition model from the storage device. The processing unit 4 inputs the word and phrase data from the word database and the important phrases corresponding to the news materials into the recognition model for analysis, thereby generating multiple keywords corresponding to the news category in the news materials. Then, the processing unit 4 identifies each keyword as either a risk keyword (e.g., but not limited to debt risk) or an industry keyword (e.g., but not limited to real estate industry). The processing unit 4 generates multiple public opinion information messages based on any one of the risk keywords and any one of the industry keywords. Each public opinion information message includes a key event and a quantity corresponding to the key event. The processing unit 4 combines the risk keyword and the industry keyword to generate the key event (e.g., but not limited to "real estate industry_debt risk"). The processing unit 4 compares the news materials included in the news category based on the risk keyword and the industry keyword to calculate the quantity of news materials containing the risk keyword and the industry keyword. When the processing unit 4 compares and finds that any news material contains risk keywords and industry keywords, the processing unit 4 determines that the news material is a target material, and the processing unit 4 counts the target material to generate the number corresponding to the key event.

[0025] Furthermore, the processing unit 4 reads the threshold from itself and compares the quantity of each piece of public opinion information with the threshold. When the processing unit 4 determines that the quantity is greater than the threshold, it generates a risk list based on the value of the public opinion information and the key event corresponding to that value. The processing unit 4 then transmits the risk list to the internal server 9 via the communication unit 2, so that the bank's processing personnel can perform relevant operations as quickly as possible based on the risk list.

[0026] In addition, the processing unit 4 will download multiple pieces of industry data from the internal server 9 via the communication unit 2. Each piece of industry data includes an industry category and an industry name. Then, the processing unit 4 checks the industry data against the industry keywords. When the processing unit 4 finds that one of the industry categories and the industry name of one piece of industry data is the same as the industry keyword, for example, the industry name and the industry keyword are both "real estate", the processing unit 4 defines the industry data as a risk industry data, and the processing unit 4 sends the risk industry data back to the internal server 9, so that the bank's processing personnel can perform subsequent processing through the risk industry data and the risk list.

[0027] In summary, through the above process, the corporate public opinion detection system 1 of the present invention can download the news data at the fixed time, automatically classify the news data into news categories, generate important phrases based on the natural language model, and then generate public opinion data and risk list based on the keywords generated by analyzing the word data and important phrases based on the recognition model, and send them back to the internal server 9 for processing personnel to use. It can improve the efficiency of processing, prevent the failure to handle events in time when they escalate, and reduce the risk of possible human misjudgment and oversight, thereby solving the problem that the previous technology was not effective when processed manually.

[0028] However, the above description is only an embodiment of the present invention and should not be construed as limiting the scope of the present invention. Any simple equivalent changes and modifications made in accordance with the scope of the patent application and the contents of the patent specification shall still fall within the scope of the patent of the present invention. [Simplified Explanation of the Diagram]

[0014] Other features and effects of the present invention will be clearly presented in the embodiments with reference to the drawings, wherein: Figure 1 is a block diagram illustrating an embodiment of the corporate public opinion detection system of the present invention.

Claims

1. A corporate public opinion detection system, applicable to a data source server, and comprising: a communication unit for providing networking functionality to communicate with the data source server; a storage unit storing a natural language model, a recognition model, a word database, and a threshold; and a processing unit electrically connected to the communication unit and the storage unit, wherein... The processing unit downloads multiple news articles from the data source server via the communication unit, generates multiple preset phrases for each news article, and categorizes the news articles into at least one news category containing the news articles using these preset phrases. The processing unit then analyzes the news articles within each news category using a natural language model to generate multiple important phrases corresponding to each news article. The processing unit reads the word database from the storage device and inputs multiple word and phrase data from the word database, along with the important phrases corresponding to the news articles, into the recognition model for analysis, generating multiple keywords corresponding to the news category within the news articles. The processing unit identifies each keyword as either a risk keyword or an industry keyword, and generates multiple public opinion information messages based on these keywords. Each public opinion information message includes a key event and a corresponding number of events. The processing unit generates public opinion information based on any of the risk keywords and any of the industry keywords. The processing unit combines the risk keywords and the industry keywords to generate the key event. The processing unit compares the news data contained in the news category based on the risk keywords and the industry keywords. When the processing unit finds that any news data contains both risk keywords and industry keywords, the processing unit determines that the news data is a target data. The processing unit counts the target data to generate the number corresponding to the key event. The processing unit compares the number of each piece of public opinion information with the threshold. When the processing unit determines that the value of the number is greater than the threshold, the processing unit generates a risk list based on the number of public opinion information and the key event corresponding to the number.

2. The enterprise public opinion detection system as described in claim 1, wherein, The storage unit also stores an exclusion list containing multiple specific words, and the processing unit checks the news materials against the specific words; when the processing unit finds that any news material contains any of the specific words, the processing unit excludes the news material.

3. The enterprise public opinion detection system as described in claim 2, wherein, The processing unit uses the natural language model to determine the various parts of speech and frequencies corresponding to these important phrases.

4. The enterprise public opinion detection system as described in claim 3, wherein, The news materials also contain multiple publication times. The processing unit determines a weight data corresponding to each news material based on the publication time of each news material and a current time, and the processing unit adds the weight data to the corresponding news material.

5. The enterprise public opinion detection system as described in claim 4 is also applicable to an internal server of a bank, the internal server being communicatively connected to the communication unit; when the processing unit generates the risk list, the processing unit transmits the risk list to the internal server via the communication unit.

6. The enterprise public opinion detection system as described in claim 5, wherein, The processing unit downloads multiple industry data from the internal server via the communication unit. Each industry data includes an industry category and an industry name. The processing unit checks the industry data against the industry keywords. When the processing unit finds that one of the industry category and the industry name of one of the industry data is the same as the industry keyword, the processing unit defines the industry data as a risky industry data and sends the risky industry data back to the internal server.

7. The enterprise public opinion detection system as described in claim 6, wherein, When the processing unit determines the weight data based on each piece of news information, the value of the weight data is inversely proportional to the difference between the publication time and the current time.

Images