KR20220093671A - An analysis appartus for social network service based on artificial intelligence - Google Patents

Abstract

According to one embodiment of the present invention, an apparatus for analyzing a social network service based on artificial intelligence (AI) includes: a data collection module collecting data on a social network service platform; a data management module outputting the data collected by the data collection module based on a pre-set search algorithm, and having a function of generating labeling for the collected data; a machine learning module performing machine learning based on a label of the data stored in the data management module, and generating an algorithm automatically performing labeling of the data; a user service module receiving a request of a user and visualizing and outputting a result analyzed by at least one between the data management module and the machine learning module based on the request of the user; and a core service module controlling at least one among the data collection module, the data management module, the machine learning module, and the user service module. The apparatus for analyzing a social network service based on AI of the present invention can be easily applied regardless of user's environment since each module is provided as a container which can operate without being affected by the environment.

Description

AI-based social network service analysis device {AN ANALYSIS APPARTUS FOR SOCIAL NETWORK SERVICE BASED ON ARTIFICIAL INTELLIGENCE}

The present invention relates to a social network service analysis device, and more specifically, to an artificial intelligence-based social network service analysis device capable of analyzing the positive and negative levels of customers by analyzing posts and comments on Internet news and social network services. it's about

Social data is in the spotlight as a marketing indicator that can best understand consumers because it is data that reflects consumer opinions or emotions, rather than processed data like survey or interview data.

Accordingly, sentiment analysis, which automatically analyzes the consumer's opinion on the target in social data or the emotion the consumer feels from the target, is also attracting attention as one of the core technologies of the big data environment.

Sentiment analysis, also called opinion mining, is a natural language processing technology that extracts people's subjective data from various contents and analyzes them to derive people's opinions and emotions.

Sentiment analysis is largely divided into a document/sentence unit sentiment analysis technique and an attribute unit sentiment analysis technique. Among them, the document unit sentiment analysis technique is mainly performed through a machine learning model.

For example, the document-level sentiment analysis technique is performed by inferring the sentiment class (e.g. positive, negative) of a given document through the learned machine learning model.

In the conventional case, data collection, data management, machine learning, etc. were performed in one system in order to perform the above-described sentiment analysis, and thus there has been a problem in that the function cannot be expanded and the user environment is affected.

On the other hand, the following prior art literature is a sentiment that enables accurate analysis and prediction of causality with variables representing macro trends in the market by calculating article polarity and constructing a monthly emotional index for each topic based on the sentiment dictionary construction for each topic and sentiment dictionary. Disclosed is a system and method for analyzing housing market trends through prior construction, but the technical gist of the present invention is not disclosed.

Republic of Korea Patent Publication No. 10-2020-0022144

AI-based news and social network service analysis apparatus according to an embodiment of the present invention aims to solve the following problems in order to solve the above problems.

It is to provide an AI-based news and social network service analysis device that can utilize the server's resources for each function to the maximum by configuring the module responsible for each function of the analysis device to be provided by each server.

In addition, it is to provide an AI-based news and social network service analysis device that can be easily applied regardless of the user's environment by setting and providing the environment of each module as an environment independent of each other.

The problems to be solved of the present invention are not limited to those mentioned above, and other problems not mentioned will be clearly understood by those of ordinary skill in the art from the following description.

An AI-based social network service analysis apparatus according to an embodiment of the present invention includes: a data collection module for collecting data on a social network service platform;

a data management module having a function of outputting the data collected by the data collection module based on a preset search algorithm and generating a label on the collected data; a machine learning module that performs machine learning based on the label of the data stored in the data management module, and generates an algorithm for automatically performing the labeling of the data; a user service module that receives a user's request, and visualizes and outputs a result analyzed by at least one of the data management module and the machine learning module based on the user's request; and a core service module for controlling at least one of the data collection module, data management module, machine learning module, and user service module.

The data collection module may include: a URL collection unit configured to collect URLs of posts posted on the social network service platform at preset intervals; and a crawling unit configured to crawl the post based on the URL received from the URL collection unit.

The URL collecting unit accesses the social network service platform at preset intervals to obtain IDs of posts from the latest post of the social network service platform to the last cycle, and then retrieves the URLs of the posts based on the IDs of the posts. It is preferable to generate and transmit to the crawling unit, and the crawling unit accesses the URL generated by the URL collection unit and then crawls the post of the corresponding URL.

Preferably, the crawling unit accesses the URL generated by the URL collection unit, crawls the posts of the URL, and then performs pre-processing to generate pre-processing data, and the pre-processing data is stored in the second database of the core service module. do.

The data management module is configured to, based on a user's request inputted from the user service module, store target data corresponding to the request among data stored in the second database of the core service module and information on the target data in the data management module. 1 It is desirable to store it in the database.

The data management module outputs the post stored in the first database on a data management web page, and on the data management web page, at least one of the original text of the post, the context in which the post is divided into preset units, and the comment of the post It is preferable that one is output.

On the data management web page, the user allows the user to label the main element (Aspect) for at least one of the original text of the post, the context in which the post is divided into preset units, and the comment of the post and the emotional evaluation of the main element Preferably, the labeling result is stored in at least one of the first database and the second database, and the preset unit is a unit that can best express the context of the corresponding post.

The AI-based social network service analysis apparatus according to an embodiment of the present invention can be easily applied regardless of the user's environment because each module is provided as a container that can operate independently of the environment, and furthermore, the performance and function of each container has the effect of being easy to expand.

Effects of the present invention are not limited to those mentioned above, and other effects not mentioned will be clearly understood by those of ordinary skill in the art from the following description.

1 is a block diagram illustrating an AI-based social network service analysis apparatus according to an embodiment of the present invention.
2 is a diagram for explaining an operation of a data collection module in an AI-based social network service analysis apparatus according to an embodiment of the present invention.
3 is a diagram for explaining an operation of a data management module in an AI-based social network service analysis apparatus according to an embodiment of the present invention.
4 and 5 are diagrams illustrating a data management web page in an AI-based social network service analysis apparatus according to an embodiment of the present invention.
6 is a view for explaining the operation of another embodiment of the data management module of the AI-based social network service analysis apparatus according to an embodiment of the present invention.
7 is a diagram illustrating a labeling storage process created by a user in a data management module of an AI-based social network service analysis apparatus according to an embodiment of the present invention.
8 is a diagram illustrating an embodiment of a screen on the Web of a machine learning module among AI-based social network service analysis apparatuses according to an embodiment of the present invention.
9 is a diagram for explaining a learning process in a machine learning module among the AI-based social network service analysis apparatus according to an embodiment of the present invention.
10 is a diagram illustrating another embodiment of a screen on the Web of a machine learning module among the AI-based social network service analysis apparatus according to an embodiment of the present invention.
11 is a diagram for explaining an inference process in a machine learning module among the AI-based social network service analysis apparatus according to an embodiment of the present invention. 12 and 13 are diagrams illustrating examples of visualization by a user service module among AI-based social network service analysis apparatuses according to an embodiment of the present invention.

A preferred embodiment according to the present invention will be described in detail with reference to the accompanying drawings, but the same or similar components are assigned the same reference numerals regardless of reference numerals, and redundant description thereof will be omitted.

In addition, in the description of the present invention, if it is determined that a detailed description of a related known technology may obscure the gist of the present invention, the detailed description thereof will be omitted. In addition, it should be noted that the accompanying drawings are only for easy understanding of the spirit of the present invention, and should not be construed as limiting the spirit of the present invention by the accompanying drawings.

Hereinafter, an AI-based news and social network service analysis apparatus according to an embodiment of the present invention will be described with reference to FIGS. 1 to 8 .

An AI-based news and social network service analysis device according to an embodiment of the present invention relates to an analysis device capable of identifying positive and negative levels of customers by analyzing posts on Internet news and social network services and comments on posts. , is configured to include a data collection module 100 , a data management module 200 , a machine learning module 300 , a user service module 400 and a core service module 500 as shown in FIG. 1 .

Each of the components constituting the AI-based social network service analysis apparatus according to an embodiment of the present invention described above is configured to include a respective server, and through this, is provided so that the server's resources can be maximally utilized for each function.

In addition, since each configuration is composed of a container-type independent environment, and there is no environmental dependence between each configuration, the configuration of the development environment is easy to optimize, and it can be easily applied regardless of the user's environment, and the performance and function of each container is configured to facilitate the expansion of

The data collection module 100 performs a function of collecting data on social network service platforms such as Naver articles, Naver cafes, Twitter, Facebook and Instagram, and has two components per one social network service platform, namely, It is configured to include a URL collection unit 110 and a crawling unit 120 .

The URL collection unit 110 performs a function of collecting URLs of articles of a specific period according to a preset period, and the crawling unit 120 receives the URL collected by the URL collection unit 110, actually crawls it, and stores it in the DB. perform the function

Whenever the URL collection unit 110 collects URLs, it transmits a queue containing URL information to the crawling unit 120, and the crawling unit 120 receives the queue sent by the URL collection unit 110 and crawls immediately. Queues that have not been processed due to the difference in operation time between the URL collection unit 110 and the crawling unit 120 are accumulated in the crawling unit 120, and data is collected while sequentially processing them.

In this case, data is periodically crawled in the background even if the user does not care, and if the cycle is shortened, real-time data collection is possible. Accurate and meaningful analysis is possible because data can be collected efficiently and data can be kept up-to-date at all times.

Hereinafter, the operation of the above-described data collection module 100 will be described in time series with reference to FIG. 2 .

First, the URL collecting unit 110 is executed according to a preset cycle to access the corresponding social network platform (S110), and then the URL collecting unit 110 is performed from the latest posting from the corresponding social network platform to the last cycle. A step (S120) of obtaining IDs of a post of

Thereafter, the URL collection unit 110 generates a URL for each of the obtained IDs through a looping statement (S130) and a step (S140) of transmitting the generated URLs to the crawling unit 120 is performed.

Next, the crawling unit 120 accesses the received URL ( S150 ) and performs pre-processing such as deleting html tags after crawling the articles of the URL ( S160 ).

After that, the crawling unit 120 stores the preprocessed data in the first database of the core service module 500 ( S170 ), and finally, the core service module 500 sends the crawling unit 120 to the data in the database. A step (S180) of transmitting a confirmation message for checking whether the storage of is well proceeds.

In addition, in the conventional data crawling process, an html-based paser is used to fetch URL information, but the data collection module 100 of the AI-based social network service analysis apparatus according to an embodiment of the present invention uses an lxml-based paser. By doing so, the effect of improving the collection speed compared to the prior art can be expected.

The data management module 200 is a configuration provided so that a user can efficiently access and manage the data collected by the data collection module 100, and the data collected by the data collection module 100 is based on a preset search algorithm. and output, and further configured to have a function of generating a label on the collected data.

The data management module 200 is provided on the web and has a first database 210 that is its own database, so a value to be stored in the first database 210 is retrieved through a request, and data stored in the first database 210 is retrieved. There are categories and search algorithms through , so users can select and view data.

That is, basic information of data such as source, date, keyword, and company can be categorized, and detailed information is configured to be easily sorted and accessed through search.

When the operation of the data management module 200 is described in time series with reference to FIG. 3 , first, the creation of a category for a target to be analyzed through the web page of the data management module 200 is performed by the core service module ( 500) is requested (S210) is performed.

At this time, the core service module 500 fetches data about the target, that is, crawled and stored data from the second database 510 it owns, and performs pagination. At this time, the requested URL information and page information will be saved together for use in the web URL where the data will be displayed later.

Thereafter, the core service module 500 transmits the total number of data and page information to the data management module 200 ( S220 ).

At this time, whenever the data transferred to the data management module 200 is stored in the first database 210 provided in the data management module 200, as shown in FIG. 4 , how much data has been retrieved is displayed on the web page. In this case, since the data is delivered in real time, the user can perform various tasks on the data management web page even if 100% of the data is not delivered.

In addition, the core service module 500 transmits the total number of data and page information to the data management module 200 while the core service module 500 transmits the URL information and page information stored in the database 510 to the dm_generator ( S230) is performed.

Afterwards, the dm_generator requests the actual paginated data to the core service module 500 ( S240 ) and the core service module 500 transmits the requested data to the dm_generator ( S250 ). At this time, the S240 and S250 are repeated until the number is filled.

When the dm_generator acquires the entire data of one page by repeating steps S240 and S250, the step of sending the data to the dm_creator (S260) is performed.

Next, the dm_creator transfers the data acquired from the dm_generator to the data management module 200 and stores the data in the first database 210 provided in the data management module 200 ( S270 ). At this time, the data management Information stored in the URL page designated in S230 is provided to the web page of the module 200 .

In addition, since it takes a long time for the list to appear on the web screen when all data is fetched from the above-described data management module 200 at once, the article per page is first displayed and when the edit of the text is pressed, the comment or context unit that was tied to the text It is also possible to increase the efficiency by implementing the articles to be imported together.

Meanwhile, the data management module 200 may be configured to use a full search text-based algorithm instead of the existing DB search when searching for and extracting data such as generation and search of categories other than data import.

The full search text-based search is very fast because the word is stored in fragments (inverted), and a significant improvement in speed can be achieved. Contrasting meaningful results can be obtained.

When the operation of the data management module 200 including the full search text-based search described above will be described in time series with reference to FIG. 6 , steps S210 to S270 are the same as described above, and after step S270, the data management module 200 ) is obtained through step S260, and then the data stored in the first database 210 is transferred to the Full Text search engine 220 (S280).

Thereafter, the data management module 200 transmits a message to the dm_creator indicating that data storage in the Full Text search engine 220 has been normally performed or, if it is not normal, sending a message requesting re-execution of S270 (S290). ) is performed.

On the other hand, the data management web page of the data management module 200 may be output as shown in FIG. 5 , and on the data management web page, the user allows the user to divide the original text of the post and the post into preset units. and an aspect of at least one of the comments of the post and may be configured to perform labeling for the emotional evaluation of the main element.

In particular, a user can modify data on a web page, and there is a detailed page for each (specifically collected data) article), and when the user performs an article here, these modifications are also displayed in the first database 210 in real time. It is configured to be reflected and stored.

In addition, the user can create labeling for performing machine learning of data on a web page, and the entire post can be cut into units such as paragraphs or sentences suitable for understanding context, and data can be saved, and labeling is performed for each unit can be

In addition, if there is a comment, since the comment is often related to the context of the text, it is preferable that the user can label the comment while viewing the text. It is desirable to be configured to be reflected in an algorithm to process data.

That is, the information directly labeled by the user is stored in the meta field of each post (body, context unit, comment) stored in the first database 210 and used for later learning. It is preferable to maximize the utility by storing it in a jsonb type to be able to respond to any format because the format may be different, and this labeling information is also stored in the second database 510 of the core service module 500. can

Meanwhile, it may be configured so that a user can add a custom category based on a keyword according to an embodiment of the present invention.

That is, a user can directly designate a category name and create a category, and a desired category can be created by adding or deleting keywords in the corresponding category.

A plurality of keywords can be entered at once by separating them with a comma (,), and each keyword is saved as a badge.

In addition, keywords are retrieved through a morpheme analysis-based search algorithm, and for more accurate searches, it is desirable that the added keywords are stored in the user's dictionary for accurate tokenizing.

Furthermore, a status was added during the labeling process, and when a label is created, the 'Requires Inspection' status is assigned and the administrator confirms the inspection later. Unwritten articles are stored in the labeling required state and can be organized so that they can be viewed and collected through categories.

In addition, in the labeling process, you can write a memo for each post, comment, and context, and if you press Save, it will be saved with the label, and the post will be marked with a 'Memo' badge and the memo can be added to the bulletin board at the bottom. It is also possible that the lower bulletin board is configured so that you can comment on the memo, and the memo badge can be configured so that you can delete the memo or leave a memo and remove it through the close issue button.

A time-series description of the labeling storage process created by the user described above with reference to FIG. 7 is as follows.

First, when the user writes and stores labeling on the data management module 200, the labeling information added on the first database 210 of the data management module 200 is updated and the ID is sent to the Full Text search engine 220 at the same time. and a step (S291) in which the state information is transmitted and the labeling writing state is updated.

Thereafter, the data management module 200 transmits the updated labeling information and meta information (id, text, context, comment, etc.) for the labeled post to the core service module 400 (S292) and the core service module A step (S293) of updating the labeling information by finding a post corresponding to the meta information transmitted from the second database 510 (500) is performed.

Afterwards, the core service module 500 transmits a message indicating that data transmission to the data management module 200 has been normally performed, or if it is not normal, a step S294 of requesting retransmission of data is performed. The machine learning module 300 performs machine learning based on the label of the data stored in the data management module 200 to perform a function of generating an algorithm that automatically performs the labeling of the data, that is, the first database 210 or the second database 510 performs a function of storing the inferred label in the output_meta field by learning with data of the meta field stored in at least one of the database 510 .

In the machine learning module 300, learning is carried out with all the labels and posts (text, unit, and comment) stored in the meta field, and when the user wants a specialized learning model for a specific part, the user service module 400 It is also possible to implement so that the user can designate parameters (target to be analyzed, period, keyword, etc.) in

On the other hand, a model that can be applied in the machine learning module 300 is a model using the Utilizing BERT for Aspect-Based Sentiment Analysis via Constructing Auxiliary Sentence paper. ), it is possible to predict whether a target (company) will receive a Sentiment (positive, neutral or negative) evaluation in Aspect (6 evaluation indicators) in a post by borrowing the concept of predicting whether it will be evaluated.

In addition, since the existing model used in this paper was an English-only model, the KoElectra model, which had the best performance as a result of self-test among several open Korean models, can be used. The model itself can change depending on the task the system needs to solve.

In addition, by applying the technology contained in the paper of Longformer: The Long-Document Transformer, it was implemented so that learning can take place up to 4096 tokens beyond the 512 token sequence limit of the Bert-Model used in the existing paper. It allowed learning to proceed more freely including contextual information down to the document level.

Currently, there is no official Korean Longformer pretrained model, and a model with excellent performance among the pretrained models converted for Longformer from several other published Korean pretrained models through research and experiments by the applicant is used. We are using a more memory efficient model.

However, in terms of performance, Longformer is inferior to the existing model in terms of performance, so a 2-way model learning method is used that uses the existing model for less than 512 tokens and the Longformer model for more than 512 tokens.

In particular, the machine learning module 300 may be configured such that a web is added, and learning and prediction made in the machine learning module 300 are performed on a web connected to the machine learning module 300 .

In such a web, a user can learn and predict with default settings, and as shown in FIG. 8, can directly set options configured for a desired task to proceed with learning. It can be configured to

In addition, the user can specify a name for the model, so that the trained model can be efficiently managed in various ways, and the model can be selected and predicted with the name that was specified directly in the inference job.

When learning is performed in the machine learning module 300, the model name, accuracy, creation date, etc. are stored in the third database 310 provided in the machine learning module 300, and related information can be checked in the model list of the machine learning module web. be able to

In particular, a process in which learning occurs in the machine learning module 300 will be described in time series with reference to FIG. 9 as follows.

First, when the user presses the learning start icon on the screen of FIG. 8 in the model learning tab of the WEB connected to the machine learning module 300, a step (S311) of entering the option information set by the user into the internal learning module is performed.

Thereafter, the machine learning module 300 transmits the transmitted information to the training worker and at the same time stores the option information in the third database 310 connected to the machine learning module 300 ( S312 ) is performed, and through this You can check the information and status of the learning model on the model list page.

After that, the training worker directly accesses the second database 510 of the core service module 500 and searches for the option information received in step S312 (S313), the second database 510 is the data (learning) retrieved in step S313. A step (S314) of delivering the necessary text, meta (information such as a label written by a person) to the training worker (S314) and the step (S315) of the training worker delivering the information received in step S314 to the machine learning module 300 are sequential proceeds with

Thereafter, the machine learning module 300 proceeds learning based on the information received in step S315 and stores the model on which the learning is completed (S316) and the state at the same time as storing the model Path on which the learning is completed in the third database 310 A step (S317) of updating to completion of learning during learning is performed. On the other hand, the model learned by the machine learning module 300 predicts the result for new data without the need for a person to directly perform labeling. Predicting the label for sentiment may be stored in at least one of the second database 510 and the third database 310, which is preferably stored to be distinguished from the label directly input by a person.

The prediction process of the machine learning module 300 will be described with reference to FIGS. 10 and 11 as follows.

First, when the user presses the prediction start icon on the web of the machine learning module 300 shown in FIG. 10, a step (S321) of transferring pre-written option information to the machine learning module is performed, and then the machine learning module 300 is S321 A step S322 of transferring the information transmitted in the step to the Inference Worker and the third database 310 is performed, and at this time, the state of the corresponding model in the third database 310 is updated to be predicted.

After the inference worker requests data corresponding to the information obtained in step S322 from the second database 510 (S323) and the second database 510 delivers the requested data to the inference worker (S324) proceeds sequentially.

Thereafter, the inference worker transmits the received data to the machine learning module 300 ( S325 ) and the machine learning module 300 performs inference based on the learned model ( S326 ).

In the machine learning module 300, a step (S327) of updating the corresponding contents in the second database 510 is performed so that the inferred data can be reflected, and at this time, the data is stored in a field named output_meta, and the meta field is the label created by the user, and output_meta is the label predicted by the model.

Finally, when the speculation on all data is completed and all updates in the second database 510 are completed, the third database 310 updates the current state to the speculation completed state ( S328 ) is performed.

When the above-described process is described in time series with reference to FIG. 6 , first, a new post crawled by the data collection module 100 , specifically, the crawling unit 120 is stored in the database of the core service module 500 ( S310 ) ) is performed, but at the same time, the data collection module 100 transmits the crawled data to the model_inference worker (S320).

Here, for higher accuracy in practice, the case that the model does not predict or the prediction score does not exceed a certain number is set as a threshold, and the prediction that does not exceed that number is a keyword developed for a project, not deep learning. By performing label prediction through a rule-based tuning algorithm, it is possible to conduct analysis through more accurate results in practice.

The user service module 400 receives a user's request, and performs a function of visualizing and outputting a result analyzed by at least one of the data management module and the machine learning module based on the user's request.

That is, the user service module 400 performs a function of analyzing the text and information stored in the database (the source of the text, date, keyword, target, aspect and sentiment) and showing the result as a diagram.

The information shown may vary depending on the task, and as shown in FIG. 12, each score by evaluation index by company (the score that considers the positive/negative evaluation of the comments on the post and how the positive/negative evaluation was performed in an aspect per each post) It can provide a radar graph (a hexagonal numerical graph), and as shown in FIG. 13, it can be configured to check the change trend of the evaluation index for each period.

In addition, the user service module 400 may provide a chart for text-based search and analysis, such as the amount of mention by platform of a specific company, word cloud, and the amount and increase rate of mention of a specific keyword, and furthermore, a trend map may also be provided. It is preferable to be configured so that

In addition, it is desirable to be implemented so that intuitive comparison is possible because different companies can be viewed overlaid on each graph. It can be implemented as a code that can show interactive results such as

In addition, it can be implemented so that various diagrams and various conditions can be set to guarantee various analyzes of the user. In the case of data used for search, if there is a meta field, the data of the meta field is used preferentially, Implement to use the model prediction value using the data of output_meta only in case.

Finally, the core service module 500 performs a function of integrated management and control of at least one of the data collection module 100, the data management module 200, the machine learning module 300, and the user service module 400, Real-time communication is made by being connected to each of the above-mentioned modules, and it plays a role of delivering the command received from each module to the appropriate module to enable the worker of the corresponding module to operate.

In addition, the core service module 500 is provided with a second database 510 in which the entire data crawled by the data collection module 100 is stored, so as to also serve as a function of delivering necessary data to each service. can play a role

An AI-based social network service analysis apparatus according to an embodiment of the present invention will be summarized and described in detail in consideration of the foregoing.

First, in the data collection module 100, a generator worker for each social platform is executed periodically for a set period, and when the generator worker is executed, the crawler worker is automatically executed and managed so that data collection can proceed.

The data management module 200 receives a request through API communication directly, delivers data information suitable for the request, and delivers actual data through dm_generator worker and dm_creator worker to operate, and internally data to inference worker through crawler worker It is managed so that even model inference is automatically transmitted.

Since all of the above systems are provided by each server, the resources of the server can be utilized to the maximum for each function, and since it is provided as a container that can operate independently of the environment that is emerging these days, it is provided in a preset environment, so it is easy regardless of the user's environment. It is applicable, and the effect that the performance and function of each container can be easily extended can be expected.

The embodiments described in this specification and the accompanying drawings are merely illustrative of some of the technical ideas included in the present invention. Therefore, since the embodiments disclosed in the present specification are for explanation rather than limitation of the technical spirit of the present invention, it is obvious that the scope of the technical spirit of the present invention is not limited by these embodiments. Modifications and specific embodiments that can be easily inferred by a person of ordinary skill in the art within the scope of the technical idea included in the specification and drawings of the present invention are included in the scope of the present invention. will have to be interpreted.

100: data acquisition module
200: data management module
300: machine learning module
400: user service module
500: core service module

Claims (7)

a data collection module for collecting data on the social network service platform;
a data management module having a function of outputting the data collected by the data collection module based on a preset search algorithm and generating a label on the collected data;
a machine learning module that performs machine learning based on the label of the data stored in the data management module, and generates an algorithm for automatically performing the labeling of the data;
a user service module that receives a user's request, and visualizes and outputs a result analyzed by at least one of the data management module and the machine learning module based on the user's request; and
a core service module for controlling at least one of the data collection module, data management module, machine learning module, and user service module;
AI-based social network service analysis device that includes.
The method according to claim 1, wherein the data collection module,
a URL collecting unit that collects URLs of posts posted on the social network service platform at preset intervals; and
a crawling unit that crawls the postings based on the URL received from the URL collection unit;
AI-based news and social network service analysis device comprising a.
3. The method according to claim 2,
The URL collecting unit accesses the social network service platform at preset intervals to obtain IDs of posts from the latest post of the social network service platform to the last cycle, and then retrieves the URLs of the posts based on the IDs of the posts. generated and delivered to the crawling unit,
The crawling unit accesses the URL generated by the URL collection unit and then crawls the post of the URL.
4. The method according to claim 3,
The crawling unit accesses the URL generated by the URL collection unit, crawls the post of the URL, and performs pre-processing to generate pre-processing data,
The AI-based social network service analysis device, characterized in that the pre-processing data is stored in the second database of the core service module.
4. The method according to claim 3,
The data management module is configured to, based on a user's request input from the user service module, store target data matching the request among data stored in the second database of the core service module and information on the target data in the data management module. 1 AI-based social network service analysis device that stores in the database.
6. The method of claim 5,
The data management module outputs the post stored in the first database on the data management web page,
An AI-based social network service analysis apparatus in which at least one of the original text of the post, the context in which the post is divided into preset units, and the comment of the post is output on the data management web page.
7. The method of claim 6,
On the data management web page, the user allows the user to label the main element (Aspect) for at least one of the original text of the post, the context in which the post is divided into preset units, and the comment of the post and the emotional evaluation of the main element is configured to perform
The AI-based social network service analysis apparatus, characterized in that the labeling result is stored in at least one of the first database and the second database.

Images