KR102575603B1 - User review analysis device by artificial intelligence and method thereof - Google Patents

Abstract

The present invention relates to an apparatus and method for analyzing a user's review of a product or service, and more particularly, to an apparatus and method for automatically analyzing user reviews using artificial intelligence (AI). it is about A review analysis device according to an embodiment of the present invention includes a processor and a memory connected to the processor and storing an artificial intelligence algorithm and review data, wherein the memory includes preset embedding of labeled learning data executable by the processor. Embedding by the embedding method, and learning the artificial intelligence algorithm using the result of inputting the embedding result into the first layer set in advance and the result of inputting the embedding result into the second layer set in advance, and using the learned artificial intelligence algorithm Thus, program instructions for classifying review data according to preset dimensions may be stored. According to the present invention, reviews written by users can be automatically multi-dimensionally analyzed using artificial intelligence, so that general users can easily decide whether to use a product or service through multi-dimensional analysis results, and sellers can easily decide whether to use a product or service. You can obtain objective information about your product or service.

Description

User review analysis device and method using artificial intelligence {USER REVIEW ANALYSIS DEVICE BY ARTIFICIAL INTELLIGENCE AND METHOD THEREOF}

The present invention relates to an apparatus and method for analyzing a user's review of a product or service, and more particularly, to an apparatus and method for automatically analyzing user reviews using artificial intelligence (AI). it is about

An artificial intelligence (AI) system is a computer system that implements human-level intelligence, and unlike existing rule-based smart systems, machines learn, judge, and become smarter on their own. The more AI systems are used, the higher the recognition rate and the more accurate understanding of user preferences. Existing rule-based smart systems are gradually being replaced by deep learning-based AI systems.

Artificial intelligence technology consists of machine learning (deep learning) and element technologies using machine learning. Machine learning is an algorithm technology that classifies/learns the characteristics of input data by itself, and element technology is a technology that uses machine learning algorithms such as deep learning to mimic the functions of the human brain, such as cognition and judgment. /It consists of technical fields such as prediction and knowledge expression.

The various fields where artificial intelligence technology is applied are as follows. Linguistic understanding is a technology for recognizing and applying/processing human language/characters, including natural language processing, machine translation, dialogue systems, question and answering, voice recognition/synthesis, and the like. Inference prediction is a technique of reasoning and predicting logically by judging information, and includes knowledge/probability-based reasoning, optimization prediction, preference-based planning, and recommendation. Knowledge expression is a technology that automatically processes human experience information into knowledge data, and includes knowledge construction (data creation/classification) and knowledge management (data utilization).

On the other hand, in an era where connectivity has become important due to technological development, the process of searching for information on products and services is becoming more important in the consumer's purchase path. Accordingly, a technology for analyzing emotions in a text (hereinafter referred to as 'review') written by a user after using a product or service has been developed. However, the current review analysis technology provides only integrated indicators such as "star ratings (scores directly given by users to products or services)" or only single-dimensional analysis of reviews is performed.

An object of the present invention is to provide a user review analysis device and method capable of multi-dimensionally analyzing user reviews using artificial intelligence.

According to one embodiment of the invention, a processor; and a memory connected to the processor and storing an artificial intelligence algorithm and review data, wherein the memory embeds labeled learning data executable by the processor in a preset embedding method, and the embedding The artificial intelligence algorithm is trained using the result of inputting the result to the first layer set in advance and the result of inputting the embedding result to the second layer set in advance, and the review data is stored in advance using the learned artificial intelligence algorithm. A review analysis apparatus for storing program instructions classified according to a set dimension is disclosed.

According to an embodiment, the labeled learning data includes a plurality of preprocessed data, the preprocessed data includes text information, dimension information, and quality information, and the dimension information may be any one of n preset quality-related items. can

According to embodiments, the embedding method may be BERT (Bidirectional Encoder Representations from Transformers), and the embedding result may correspond to a plurality of pieces of output information included in the encoding output of the BERT.

According to an embodiment, the memory stores program instructions for generating first processing information by inputting encoding output information for a CLS token among the plurality of output information to the first layer, wherein the first layer is linear. may correspond to a layer.

Depending on the embodiment, the memory may store program instructions for generating the 1-1 processing information by decoding the first processing information with a preset first function.

Depending on embodiments, the first function may correspond to a softmax function.

According to an embodiment, the memory generates a first loss value by inputting the 1-1 processing information to a preset first loss function, and uses the first loss value to calculate the artificial intelligence algorithm. Program commands to be learned can be stored.

According to an embodiment, the memory stores program instructions for generating second processing information by inputting all of the plurality of output information to the second layer, and the second layer may correspond to a CNN algorithm.

Depending on the embodiment, the memory may store program instructions for generating the 2-1 processing information by decoding the second processing information with a preset second function.

Depending on embodiments, the second function may correspond to a SOFTMAX function.

According to an embodiment, the memory generates a second loss value by inputting the 2-1 processing information to a preset second loss function, and uses the first loss value and the second loss value. It is possible to store program instructions for learning the artificial intelligence algorithm.

According to an embodiment, the memory may store program instructions for learning the artificial intelligence algorithm such that a sum of the first loss value and the second loss value becomes a minimum value.

According to an embodiment, the memory stores program instructions for generating third processing information by inputting encoding output information corresponding to text information among the plurality of output information to the first layer, wherein the first layer is linear. may correspond to a layer.

Depending on the embodiment, the memory may store program instructions for generating the 3-1 processing information by decoding the third processing information with a preset third function.

Depending on embodiments, the third function may correspond to either a conditional random fields (CRF) function or a softmax (SOFTMAX) function.

According to an embodiment, the memory generates a third loss value by inputting the 3-1 processing information to a preset third loss function, and uses the third loss value to calculate the artificial intelligence algorithm. Program commands to be learned can be stored.

According to the present invention, reviews written by users can be automatically multi-dimensionally analyzed using artificial intelligence, so that general users can easily decide whether to use a product or service through multi-dimensional analysis results, and sellers can easily decide whether to use a product or service. You can obtain objective information about your product or service.

In order to more fully understand the drawings cited in the detailed description of the present invention, a brief description of each drawing is provided.
1 is a block diagram of a review analysis apparatus according to an embodiment of the present invention.
2 is an operation flowchart for explaining the operation of the review analysis apparatus according to an embodiment of the present invention.
3 is a diagram exemplified to explain learning data according to an embodiment of the present invention.
4 is a diagram for explaining an example of artificial intelligence and a process of learning artificial intelligence according to an embodiment of the present invention.
5 and 6 are diagrams illustrating results obtained by analyzing a plurality of reviews by artificial intelligence according to an embodiment of the present invention.
7 is a flowchart of a review analysis operation according to an embodiment of the present invention.

Exemplary embodiments according to the technical spirit of the present invention are provided to more completely explain the technical spirit of the present invention to those skilled in the art, and the following embodiments are modified in various forms. It can be, the scope of the technical spirit of the present invention is not limited to the following embodiments. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the spirit of the invention to those skilled in the art.

Although the terms first, second, etc. are used in this specification to describe various members, regions, layers, regions and/or components, these members, parts, regions, layers, regions and/or components refer to these terms. It is self-evident that it should not be limited by These terms do not imply any particular order, top or bottom, or superiority or inferiority, and are used only to distinguish one member, region, region, or component from another member, region, region, or component. Accordingly, a first member, region, region, or component described in detail below may refer to a second member, region, region, or component without departing from the teachings of the technical idea of the present invention. For example, a first element may be termed a second element, and similarly, the second element may also be termed a first element, without departing from the scope of the present invention.

Unless defined otherwise, all terms used herein, including technical terms and scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which the concept of the present invention belongs. In addition, commonly used terms as defined in the dictionary should be interpreted as having a meaning consistent with what they mean in the context of the technology to which they relate, and in an overly formal sense unless explicitly defined herein. will not be interpreted.

The term 'and/or' as used herein includes each and every combination of one or more of the recited elements.

Hereinafter, embodiments according to the technical idea of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram of a review analysis apparatus according to an embodiment of the present invention.

Referring to FIG. 1 , a review analysis device 100 according to an embodiment of the present invention may include a modem 110, a processor 120, and a memory 130.

The modem 110 may be a communication modem that is electrically connected to other external devices (not shown) to enable mutual communication. In particular, the modem 110 may output data received from these external devices to the processor 120, and the processor 120 may store these data in the memory 130. For example, the modem 110 may output crawled data to the processor 120 from various servers (eg, portal site servers, etc.) connected through the Internet.

Hereinafter, among data crawled or downloaded from various servers, the data used to learn the 'artificial intelligence algorithm' stored in the memory 130 is referred to as "learning data", and through the learned artificial intelligence algorithm, any product or service The data used to measure 'quality by dimension' is called "review data". The training data and/or the review data may correspond to text data uploaded by one or more users to a server on the Internet.

The memory 130 is a component in which various information and program commands for the operation of the review analysis device 100 are stored, and may be a storage device such as a hard disk or a solid state drive (SSD). In particular, the memory 130 may store one or more learning data and/or review data input from the modem 110 under the control of the processor 120 . In addition, the memory 130 may store program instructions such as an artificial intelligence algorithm for executing a 'review analysis method' executable by the processor 120 .

The processor 120 may process learning data and/or review data using information and program instructions stored in the memory 130, and measure depression corresponding to the quality of a product and/or service by dimension.

Here, 'dimension' may mean an individual element for measuring the quality of a product or service in various aspects. For example, a dimension corresponding to an arbitrary restaurant could be set to include the following items.

- Price

- taste

- Kindness level

- degree of cleanliness

- atmosphere

In addition, 'quality' may be set in stages or set to positive/negative/neutral. Hereinafter, the operation of the processor 120 will be described in detail with reference to FIGS. 2 to 4 .

Figure 2 is an operation flow chart for explaining the operation of the review analysis device according to an embodiment of the present invention, Figure 3 is a diagram exemplified for explaining learning data according to an embodiment of the present invention, Figure 4 is this It is a diagram for explaining an example of artificial intelligence and a process of learning artificial intelligence according to an embodiment of the invention.

2, the processor 120 of the review analysis device 100 according to an embodiment of the present invention preprocesses learning data to generate preprocessed data, uses the preprocessed data to learn an artificial intelligence algorithm, and learns artificial intelligence. An operational flow for measuring dimensionally quality of a product or service using an intelligent algorithm is shown. Operations to be described below may be operations generated by the processor 120 executing program instructions stored in the memory 130, but will be described as being performed by the processor 120 for convenience of understanding and description.

First, the processor 120 may collect and pre-process learning data (210 and 220). For example, the processor 120 collects learning data by crawling reviews of any product or service uploaded to an Internet server, and then uses program instructions stored in the memory 130 to send the modem 110 A plurality of input learning data may be pre-processed through. Here, the preprocessing may be one or more of the following operations.

(1) Operation in which dimension information corresponding to learning data is given

(2) Operation in which quality information corresponding to the assigned dimension is assigned

(3) When the data for learning is composed of multiple sentences, the operation classified for each sentence

The operation of pre-processing the learning data may be performed automatically according to the execution of program instructions by the processor 120 or manually according to the manager's manipulation of the review analysis device 100 .

3 illustrates a case in which a plurality of labeled preprocessing data is generated using one training data.

Referring to Figure 3, a single learning data is as follows.

"The food was good, the restaurant was clean and adequate."

Pre-processing data may include text information, dimension information, and/or quality information. Here, the text information may be information corresponding to text included in the learning data. When a plurality of sentences are included in the training data, the preprocessing data may include a plurality of text information (here, the text information may correspond to short text). In addition, the dimension information is information about a dimension preset in the learning data, and the preprocessing data may include any one of a plurality of dimensions. Also, the quality information may be information about quality corresponding to a dimension.

Therefore, the learning data may be preprocessed to include the following configuration.

(1) Answer information (Yes=1 or No=0)

(2) Text information (the food was good, the restaurant was neat and appropriate)

(3) Dimensional information (restaurant_general, food_quality, restaurant_price, atmosphere_general, etc.)

(4) Quality information (positive, negative or neutral)

(5) target information (in the example of FIG. 3, food, restaurant, atmosphere, etc.)

As illustrated in FIG. 3 , the preprocessing data may include text information, one of a plurality of dimension information, quality information for each dimension, target information corresponding to the dimension, and correct answer information.

Accordingly, the processor 120 may perform supervised learning of the artificial intelligence algorithm stored in the memory 130 using the preprocessing data (Artificial Intelligent Modeling, 230).

In this case, the artificial intelligence algorithm may mean including the following program commands.

(1) A program related to language analysis (for example, BERT, Bidirectional Encoder Representations from Transformers) capable of embedding preprocessing data in a preset embedding method (hereinafter referred to as 'embedding program').

(2) A first layer (eg, a linear layer, linear layer) and/or a second layer (eg, a CNN layer, a convolutional neural layer) that processes a result embedded in an embedding program according to a preset method.

(3) one or more loss functions for generating loss values using results processed in the first layer and/or the second layer

(4) Program instructions for learning the embedding program, the first layer, the second layer, etc. so that the loss value is minimized

Referring to FIG. 4, an artificial intelligence algorithm in the case where the embedding program is BERT, the first layer is a linear layer, and the second layer is a CNN layer is exemplified. Hereinafter, an operation that an artificial intelligence algorithm learns through preprocessing data will be described.

First, preprocessing data can be tokenized in the following format to correspond to BERT, an embedding program.

"[CLS] text information [SEP] dimension information sentiment information [SEP]"

In addition, preprocessing data can be embedded (encoded) by BERT. A result embedded by BERT (hereinafter referred to as 'embedding result') may correspond to a plurality of token information included in the encoding output of BERT. The encoding output 410 of BERT corresponds to the last hidden state of tokenized preprocessing data and may include the following information.

- T[CLS] (420): encoding output information for the CLS token of BERT

- T[1] to T[n] 430: encoding output information for each of the n words included in the text information (where n is a natural number)

- T[sep],1 to T[sep], m: When m sentences are included in the text information, encoding output information for the SEP token (provided that m is a natural number) (m=2 in FIG. 4)

The processor 120 may execute program instructions stored in the memory 130 so that the embedding result is processed by the first layer 440 and the second layer 450 which are set in advance. Here, the first layer may correspond to a linear layer, and the second layer may correspond to a convolutional neural network (CNN) algorithm.

An embedding result, that is, T[CLS] 420 among a plurality of pieces of output information 410 included in the encoding output of BERT may be input to the first layer 440 . The first layer 440 may generate first processing information by processing 'T[CLS] 420 '. For example, the first layer 440 may include a pooler and a linear layer. 'T[CLS] 420' is processed by pooling (eg, MAX pooling) in a pooler, and then linearly processed in a linear layer to obtain a first vector value (hereinafter, 'first referred to as 'processing information'). Since the operation of the first layer 440 is almost the same as the published operation after BERT embedding, a detailed description thereof will be omitted.

In addition, the embedding result, that is, all of the plurality of pieces of output information 410 included in the encoding output of BERT may be input to the second layer 450 . The second layer 450 may generate second processing information by processing 'all of the plurality of output information 410'. For example, the second layer 450 is a layer corresponding to a convolutional neural network (CNN) algorithm and may include a convolutional 2D layer and an adaptive average pooler. 'All of the plurality of output information 410' may be processed through a convolutional 2D layer and an adaptive average pooler and output as a second vector value (hereinafter referred to as 'second processing information'). Since the operation of the second layer 450 corresponds to a previously disclosed operation in relation to the CNN algorithm, a detailed description thereof will be omitted.

Meanwhile, the embedding result, that is, among the plurality of output information 410 included in the encoding output of BERT, T[1] to T[n] 430 may be input to the first layer 440, and the first layer ( 440) may generate a third vector value (hereinafter referred to as 'third processing information') by processing 'T[1] to T[n] 430'.

In addition, all of the first processing information to the third processing information may be vector values having the same size. That is, the first layer 440 and the second layer 450 may be formed such that vector values having the same size are output.

The processor 120 may execute program instructions stored in the memory 130 to decode the first through third processing information using preset functions.

For example, the first processing information may be decoded in the first function 460 . Here, the first function 460 may correspond to softmax. Hereinafter, the value output from the first function 460 is referred to as 1-1 processing information. Also, the second processing information may be decoded in the second function 470. Here, the second function 470 may also correspond to softmax. Hereinafter, the value output from the second function 470 is referred to as 2-1 processing information. Also, the third processing information may be decoded in the third function 480 . Here, the third function 480 may correspond to softmax or conditional random field (CRF). The value output from the third function 480 is referred to as 3-1 processing information.

The processor 120 executes program instructions stored in the memory 130, and processing information 1-1 to 3-1 processing information is processed by a preset loss function (Loss function, 490, 492 and 495), A first loss value to a third loss value may be generated. The loss function is a function that quantifies the difference between the actual value and the predicted value, and may be the same as the public function.

For example, the 1-1 processing information may be processed in the first loss function 490 and output as a first loss value. In addition, the 2-1 processing information may be processed in the second loss function 492 and output as a second loss value. The 3-1 processing information may be processed in the third loss function 495 and output as a third loss value.

The processor 120 may execute program instructions stored in the memory 130 so that the first loss value to the third loss value are used for learning the artificial intelligence algorithm. In this case, the first loss value and the second loss value may be added and used for learning the artificial intelligence algorithm. That is, the artificial intelligence algorithm can learn to minimize the 'value obtained by adding the first loss value and the second loss value' and the 'third loss value', respectively.

As described above, the processor 120 may execute program instructions so that the artificial intelligence algorithm is learned using the 'learning data' stored in the memory 130. Hereinafter, a review analysis operation 240 in which the processor 120 automatically classifies review data by dimension by executing an artificial intelligence algorithm stored in the memory 130 will be described.

5 and 6 are diagrams illustrating results obtained by analyzing a plurality of reviews by artificial intelligence according to an embodiment of the present invention.

The processor 120 may automatically derive the result illustrated in FIG. 5 by inputting the review data stored in the memory 130 to the learned artificial intelligence algorithm described with reference to FIGS. 1 to 4 . In other words, for the review data that "there is an atmosphere unique to this restaurant that I want to revisit," dimension information corresponding to "atmosphere_general" and quality information corresponding to "positive" can be derived, and "the restaurant is too narrow." You always have to wait for 45 minutes to an hour, but you get rewarded the moment you put the food in your mouth. can

In addition, as illustrated in FIG. 6, the processor 120 inputs the review data into the learned artificial intelligence algorithm to derive a quality graph by dimension and related keywords for any product/service as illustrated in FIG. 6. can 6 is a graph showing the result of selecting two random restaurants from the Internet portal site ( www.naver.com ) server, obtaining review data, and then inputting them into an artificial intelligence algorithm and classifying them. In this case, the processor 120 may derive a graph by scoring the derived quality for each dimension. For example, the processor 120 may score quality per dimension by the following formula.

Score = [number of positive reviews - number of negative reviews] / total number of reviews

Thereafter, the processor 120 may derive a graph as illustrated in FIG. 6 using the information scored for each dimension.

7 is a flowchart of a review analysis operation according to an embodiment of the present invention.

7 shows operations of the review and analysis apparatus 100 described with reference to FIGS. 1 to 6 according to an operation sequence. Since operations performed in each step shown in FIG. 7 have already been described in detail, a detailed description thereof will be omitted.

As described above, according to the present invention, since reviews written by users can be automatically analyzed in multiple dimensions using artificial intelligence, general users can easily determine whether to use a product or service through the multi-dimensional analysis result. It is obvious that sellers can obtain objective information about their products or services.

In the above, the present invention has been described in detail with preferred embodiments, but the present invention is not limited to the above embodiments, and various modifications and changes are made by those skilled in the art within the technical spirit and scope of the present invention. this is possible

100: review analysis device
110: modem
120: processor
130: memory

Claims (16)

processor; and
a memory connected to the processor and storing artificial intelligence algorithms and review data;
Including,
The memory is executable by the processor,
The artificial learning learned using a result of embedding labeled learning data in a preset embedding method and inputting the embedding result into a preset first layer and inputting the embedding result into a preset second layer Store program instructions for classifying the review data according to preset dimensions using an intelligent algorithm;
The labeled training data includes a plurality of preprocessing data,
The preprocessing data includes text information, dimension information and quality information,
The dimension information is any one of n preset quality-related items,
The embedding method is BERT (Bidirectional Encoder Representations from Transformers),
The embedding result corresponds to a plurality of output information included in the encoding output of the BERT,
The artificial intelligence algorithm is a result of inputting encoding output information for a CLS token from among the plurality of output information to the first layer, a result of inputting all of the plurality of output information to the second layer, and a result of inputting all of the plurality of output information to the second layer. A review analysis device that is learned using a result of inputting encoding output information corresponding to text information to the first layer.
delete delete According to claim 1,
the memory,
Store program instructions for generating first processing information by inputting encoding output information for the CLS token to the first layer;
The first layer corresponds to a linear layer, review analysis device.
According to claim 4,
the memory,
A review analysis device for storing program instructions for generating 1-1 processing information by decoding the first processing information with a preset first function.
According to claim 5,
The first function corresponds to a softmax (SOFTMAX) function, review analysis device.
According to claim 5,
the memory,
A review of generating a first loss value by inputting the 1-1 processing information into a preset first loss function and storing program instructions for learning the artificial intelligence algorithm using the first loss value analysis device
According to claim 7,
the memory,
Store program instructions for generating second processing information by inputting all of the plurality of output information to the second layer;
The second layer corresponds to the CNN algorithm, review analysis device.
According to claim 8,
the memory,
A review analysis device for storing program instructions for generating 2-1 processing information by decoding the second processing information with a preset second function.
According to claim 9,
The second function corresponds to a softmax (SOFTMAX) function, review analysis device.
According to claim 9,
the memory,
A program for generating a second loss value by inputting the 2-1 processing information into a preset second loss function, and learning the artificial intelligence algorithm using the first loss value and the second loss value Review analysis device, storing instructions
According to claim 11,
the memory,
A review analysis device for storing program instructions for learning the artificial intelligence algorithm so that a sum of the first loss value and the second loss value becomes a minimum value.
According to claim 12,
the memory,
Store program instructions for generating third processing information by inputting encoding output information corresponding to text information among the plurality of output information to the first layer,
The first layer corresponds to a linear layer, review analysis device.
According to claim 13,
the memory,
A review analysis device for storing program instructions for generating 3-1 processing information by decoding the third processing information with a preset third function.
According to claim 14,
The third function corresponds to any one of CRF (Conditional Random Fields) or Softmax (SOFTMAX) function, review analysis device.
According to claim 14,
the memory,
A review of generating a third loss value by inputting the 3-1 processing information into a preset third loss function and storing program instructions for learning the artificial intelligence algorithm using the third loss value analysis device.