Apparatus, method and recording medium storing instruction for determining recommended review of a product

Abstract

The present invention provides an apparatus comprising one or more processors and one or more memories storing commands to be executed by the one or more processors. The apparatus is configured such that, when executing the commands, the one or more processors perform the following: obtaining first information from a user's terminal regarding a first review of a product; generating a first vector corresponding to the first review based on the first information; determining, based on the first vector, one or more recommended reviews among one or more second reviews of the product that are similar to the first review; and transmitting the recommendation review information of the one or more recommended reviews to the terminal.

Description

[Technical Field]

[0001] This invention relates to a technique for determining product recommendation reviews. [Previous Technology]

[0002] Using an e-commerce service that allows users to purchase goods in an online store. Users can refer to reviews of the products to determine whether to purchase them.

[0003] Users who purchase and use the product may agree or disagree with the content of various reviews of the product. When a large number of users agree to a single review, the reliability of the product quality can be improved. [Summary of the Invention]

[0004] [The problem the invention aims to solve]

[0005] At least one embodiment of the present invention can determine recommended reviews of a product.

[0006] At least one embodiment of the present invention can identify reviews similar to user reviews of products as recommended reviews.

[0007] The technical subject matter of this invention is not limited to the technical subject matter mentioned above. Those skilled in the art can clearly understand other technical subject matter not mentioned based on the description herein. [Technical Means for Solving the Problem]

[0008] An apparatus according to one embodiment of the present invention may include: one or more processors; and one or more memory stores commands for execution by the one or more processors; and the apparatus is configured such that, when executing the commands, the one or more processors perform the following: obtaining first information from a user's terminal regarding a first review of a product; generating a first vector corresponding to the first review based on the first information; determining one or more recommended reviews similar to the first review from one or more second reviews of the product based on the first vector; and transmitting the recommendation review information of the one or more recommended reviews to the terminal.

[0009] In one embodiment, the above-mentioned device can be configured such that when the first vector is generated, one or more processors input the first information into a learning model to generate the first vector as the output of the learning model, wherein the learning model learns by converting product reviews into vectors.

[0010] In one embodiment, the learning model described above may be a model that learns by using one or more second comments written for the product as learning material.

[0011] In one embodiment, the learning model described above may be a BERT (Bidirectional Encoder Representations from Transformers) model.

[0012] In one embodiment, the above-described apparatus can be configured such that, when determining the one or more recommended reviews, the one or more processors perform the following: obtaining one or more second vectors corresponding to each of the one or more second reviews; determining the similarity between the first vector and each of the one or more second vectors; determining one or more similar vectors among the one or more second vectors based on the similarity; and determining the review among the one or more second reviews corresponding to each of the one or more similar vectors as the one or more recommended reviews.

[0013] In one embodiment, the above-described apparatus can be configured such that, when determining the similarity between the first vector and one or more second vectors, the processor determines the absolute value of the distance between the first vector and one or more second vectors as the similarity.

[0014] In one embodiment, the above-described apparatus can be configured such that the one or more processors store the first comment and the first vector in association in the one or more memories.

[0015] In one embodiment, the above-described apparatus can be configured such that, when determining the above-mentioned one or more similar vectors, the above-mentioned one or more processors determine the vectors among the above-mentioned one or more second vectors whose similarity to the above-mentioned first vector is a first threshold value or above as the above-mentioned one or more similar vectors.

[0016] In one embodiment, the above-mentioned device can be configured such that the above-mentioned one or more processors further obtain from the above-mentioned terminal a request for additional comments other than the above-mentioned one or more recommended comments; in response to the above-mentioned additional comment request, determine one or more additional comments from the comments that do not belong to the above-mentioned one or more recommended comments; and transmit the additional comment information of the above-mentioned one or more additional comments to the above-mentioned terminal.

[0017] In one embodiment, the above-described apparatus can be configured such that, when determining the one or more additional comments, the one or more processors perform the following: based on the similarity values, determine one or more additional vectors among the one or more second vectors, wherein the one or more additional vectors have a similarity with the first vector that does not reach the first threshold value and is at or above the second threshold value; determine the comments corresponding to each of the one or more additional vectors as the one or more additional comments; wherein the second threshold value is less than the first threshold value.

[0018] In one embodiment, the above-mentioned device can be configured such that the above-mentioned one or more processors are as follows: obtaining consent or dissent information from the above-mentioned terminal, the consent or dissent information indicating whether the above-mentioned user agrees with at least one commenter in the above-mentioned one or more recommended comments; based on whether the above-mentioned user agrees with the at least one comment, determining one or more comments in the above-mentioned one or more second comments that do not belong to the above-mentioned one or more recommended comments as one or more additional comments; and transmitting the additional comment information of the above-mentioned one or more additional comments to the above-mentioned terminal.

[0019] One embodiment of the present invention is implemented in a device, the device including one or more processors and one or more memories storing commands to be executed by the one or more processors. The method may include the following steps: the one or more processors obtain first information of a first review filled in by a user for a product from a user's terminal; generate a first vector corresponding to the first review based on the first information; determine one or more recommended reviews similar to the first review from one or more second reviews filled in for the product based on the first vector; and transmit the recommendation review information of the one or more recommended reviews to the terminal.

[0020] One embodiment of the present invention uses a recording medium that is a non-transitory computer-readable recording medium, which records commands that, when executed by one or more processors, cause the one or more processors to perform actions. These commands are configured such that the one or more processors: obtain first information from a user's terminal regarding a first review of a product; generate a first vector corresponding to the first review based on the first information; determine, based on the first vector, one or more recommended reviews among one or more second reviews of the product that are similar to the first review; and transmit the recommendation review information of the one or more recommended reviews to the terminal. [Effects of the Invention]

[0021] According to at least one embodiment of the present invention, recommended reviews of a product can be determined.

[0022] The effects of the technical concept of the present invention are not limited to the effects mentioned above. Those who are skilled in the art can clearly understand other effects not mentioned in the present invention.

Implementation Method

[0024] The various embodiments described herein are illustrated for the purpose of clearly explaining the technical concept of the present invention and are not intended to limit them to specific implementation methods. The technical concept of the present invention includes various modifications, equivalents, alternatives, and embodiments obtained by selectively combining all or part of the embodiments described herein. Furthermore, the scope of the technical concept of the present invention is not limited to the various embodiments presented below or their specific descriptions.

[0025] The terms used herein, including technical or scientific terms, shall have the meaning commonly understood by one of common sense in the technical field to which this invention pertains, unless otherwise defined.

[0026] The expressions used herein, such as "including," "may include," "possibly possess," "may possess," "have," and "may have," imply the presence of object features (e.g., functions, actions, or constituent elements), and do not exclude the existence of other additional features. That is, the expressions described above should be understood as open-ended terms that have the possibility of including other embodiments.

[0027] The singular expressions used herein may include the meaning of the plural form unless otherwise stated in the context, and this also applies to the singular expressions described in the claims of the invention.

[0028] The terms “first,” “second,” “first,” “second,” etc., used in this document are used to distinguish one object from another when referring to multiple objects of the same kind, unless otherwise specified in the context, and are not used to limit the order or importance of such objects.

[0029] As used herein, “A, B and C”, “A, B or C”, “A, B and / or C” or “at least one of A, B and C”, “at least one of A, B or C”, “at least one of A, B and / or C”, etc., may refer to each of the listed items or all combinations of the listed items. For example, “at least one of A or B” may refer to (1) at least one A, (2) at least one B, (3) at least one A and at least one B.

[0030] As used herein, the term "component" may refer to software or hardware components such as FPGA (Field-Programmable Gate Array) or ASIC (Application Specific Integrated Circuit). However, "component" is not limited to software or hardware. A "component" can be constructed in a manner that is located on addressable storage media or can be constructed in a manner that executes one or more processors. In one embodiment, a "component" may include components such as software components, object-oriented software components, class components and task components, processes, functions, attributes, programs, sub-formulas, code fragments, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays and variables.

[0031] The term “based on” as used herein is used to describe one or more factors that affect the determination or judgment of an action or behavior described in the statement or article that includes the term. The term does not exclude other factors that affect the determination or judgment of an action or behavior.

[0032] The expression used herein to describe a constituent element (e.g., the first constituent element) as “connected” or “linked” to another constituent element (e.g., the second constituent element) may mean that the aforementioned constituent element is directly connected to or linked to the aforementioned other constituent element, or is connected to or linked to the aforementioned other constituent element through a new other constituent element (e.g., the third constituent element).

[0033] Depending on the context, the expression "configured to" as used herein may mean "configured in a specific manner," "possessing the capability to perform specific actions," "modified in a specific manner," "made in a specific manner," or "capable of performing specific actions." This expression is not limited to the meaning of "specially designed in hardware." For example, a processor configured to perform a specific action may refer to a generic-purpose processor that can perform that specific action by executing software.

[0034] Hereinafter, various embodiments of the present invention will be described with reference to the accompanying drawings. In the drawings and descriptions thereof, the same or substantially equivalent constituent elements are given the same reference numerals. Furthermore, in the following descriptions of various embodiments, repeated descriptions of the same or corresponding constituent elements may be omitted, but this does not mean that the constituent element is not included in the embodiment.

[0035] Figure 1 is a diagram illustrating a system 100 according to various embodiments of the present invention. System 100 may include a user terminal 110 and a device 120. The user terminal 110 may refer to a terminal device used by a customer (user) who orders goods and fills out reviews. For example, the terminal device may be a portable communication device, a computer device, a portable multimedia device, a wearable device, a home appliance device, or a device composed of one or more of the above devices. The user terminal 110 may include a display device for displaying one or more reviews filled out for the goods. The display device may be a device that provides various information to the user in a visual form. For example, the display device may include a monitor, a projector, a hologram, etc. The device 120 may be a device of a service provider that provides e-commerce services for goods. The device 120 may be a server that collects reviews of goods and determines recommended reviews.

[0036] The user's terminal 110 may be a device capable of connecting to the Internet. The user's terminal 110 may connect to the device 120 via the network to send and receive various data. The user's terminal 110 may transmit first information of a first review filled out by the user regarding a product to the device 120. In this document, the first review may refer to a review filled out by the user regarding a product, and the first information refers to information including the information of the first review. The device 120 may be configured to determine a recommended review among one or more second reviews based on the first information obtained from the user, and transmit the recommended review to the user's terminal 110. In this document, the second review may refer to a review pre-filled regarding a product. The recommended review may be displayed to the user via the display device of the user's terminal 110.

[0037] FIG2 is a diagram illustrating an example of a device 120 according to an embodiment of the present invention. The device 120 according to an embodiment of the present invention may include a processor 210 and / or memory 220. At least one of the components included in the device 120 may be omitted, or other components may be added to the device 120. For example, the device 120 may further include a communication circuit 230 for communicating with an external device. The device 120 may be implemented by additionally or alternatively integrating some of its components, or implemented as a single or multiple entities. At least some of the components within the device 120 may be interconnected via buses, GPIO (General Purpose Input / Output), SPI (Serial Peripheral Interface), or MIPI (Mobile Industry Processor Interface), etc., to transmit and receive data and / or signals.

[0038] According to various embodiments, the processor 210 of the device 120 may be configured to perform control of various components of the device 120 (e.g., memory 220) and / or communication-related calculations or data processing. The processor 210 may drive software to control at least one component of the device 120 connected to the processor 210. Furthermore, the processor 210 may perform various calculations, processing, data generation, and processing operations associated with the present invention. Also, the processor 210 may load data from memory 220 or store such data in memory 220. In embodiments of the present invention, the processor 210 may execute commands stored in memory 220 to determine product recommendations and reviews. The processor 210 disclosed herein may also refer to a collection of more than one processor 210.

[0039] According to various embodiments, the memory 220 of the device 120 may store commands for actions of the processor 210. Data stored in the memory 220, as data obtained, processed, or used by at least one component of the device 120, may include software (e.g., a program). The memory 220 may include volatile and / or non-volatile memory. The memory 220 may also store information processed during the operation of the processor 210. In embodiments of the invention, the memory 220 may store information on one or more reviews written for a product and one or more vectors corresponding to each of the reviews.

[0040] According to various embodiments, the communication circuit 230 of device 120 can establish a wired or wireless communication channel with an external device (e.g., a user's terminal 110) to send and receive various data. According to one embodiment, the communication circuit 230 may include at least one port for connecting to an external device via a wired cable for wired communication. In the above case, the communication circuit 230 can communicate with a wired external device via at least one port. According to one embodiment, the communication circuit 230 can be configured to include a cellular communication module connected to a cellular network (e.g., 3G, LTE, 5G, Wibro, or WiMAX). According to various embodiments, the communication circuit 230 may include a short-range communication module to send and receive information with an external device using short-range communication (e.g., Wi-Fi, Bluetooth, Bluetooth Low Energy (BLE), UWB), but is not limited thereto. According to one embodiment, the communication circuit 230 may include a contactless communication module for contactless communication. Contactless communication may include, for example, at least one of the following short-range contactless communication technologies: NFC (Near Field Communication), RFID (Radio Frequency Identification), or MST (Magnetic Secure Transmission).

[0041] FIG3 is a sequence diagram illustrating the operation method of the apparatus 120 according to various embodiments of the present invention. In various embodiments, the memory 220 may store commands to be executed in the processor 210 and information processed in the processor 210, which may perform the following operation method by executing commands.

[0042] In step S310, the processor 210 can obtain the first information of the first review filled in by the user regarding the product from the user's terminal 110. In various embodiments, the processor 210 can receive the first information from the user's terminal 110 via the communication circuit 230. In one example, the processor 210 can directly obtain the first information received by the device 120 via the communication circuit 230. In another example, the first information can be temporarily stored in the memory 220, and the processor 210 can also access the memory 220 to obtain the first information.

[0043] In step S320, processor 210 may generate a first vector corresponding to the first comment based on the first information. The first comment may include text. In various embodiments, processor 210 may convert the text of the first comment into a vector to generate the first vector.

[0044] In step S330, the processor 210 may determine, based on the first vector, one or more recommended reviews among one or more second reviews for a product that are similar to the first review. The second review may include text. In various embodiments, the processor 210 may obtain one or more second vectors that convert the text of each of the one or more second reviews into vectors. In one example, memory 220 may store one or more second vectors, and the processor 210 may obtain one or more second vectors from memory 220. In another example, one or more second vectors may be distributed and stored in one or more external server devices, and the processor 210 may request one or more second vectors from one or more external server devices via communication circuit 230, and obtain one or more second vectors in response to the request. In various embodiments, the processor 210 may determine the similarity between the first vector and one or more second vectors, and determine one or more recommended reviews based on the similarity.

[0045] In step S340, the processor 210 may transmit recommendation comment information of one or more recommended comments to the terminal. In various embodiments, the processor 210 may transmit the recommendation comment information to the user's terminal 110 via the communication circuit 230. The user's terminal 110 may display the recommended comments to the user via a display device based on the recommendation comment information.

[0046] Figure 4 is a sequence diagram illustrating an example of generating the first vector in step S320 of Figure 3. In step S410, the processor 210 may use one or more second comments as learning data to train the learning model. In various embodiments, the learning model may be an artificial intelligence model that outputs the input text as a vector. The learning model may be a model based on a transformer architecture. For example, the learning model may be based on the BERT (Bidirectional Encoder Representations from Transformers) model.

[0047] In step S420, processor 210 may generate a first vector by inputting first information into a learning model. In various embodiments, processor 210 may be adapted to preprocess the first information in a manner that is input into a learning model to generate first preprocessed information. Processor 210 may input the first preprocessed information into a learning model to generate a first vector as the output of the learning model. Processor 210 may store the first vector in memory 220 in association with a first comment.

[0048] Figure 5 is a diagram illustrating an example of training the learning model using the second review in step S410 of Figure 4. In various embodiments, memory 220 may store commands for execution in processor 210, by which processor 210 may perform the following actions. Processor 210 may be configured to train the learning model using one or more second reviews stored in memory 220 as learning data.

[0049] In various embodiments, when training the learning model, the processor 210 may utilize a pre-learning model 510 that has pre-learned the language features of one or more second comments. For example, the pre-learning model 510 may be one that learns the language features of the input sentence in a way that understands the meaning and features of the input sentence.

[0050] In one embodiment, where one or more second comments are written in Korean, the processor 210 can load a pre-learning model 510, which has pre-learned the linguistic features of the Korean text, from memory 220. As an example, the pre-learning model 510 can be an Albert (albert-kor-base) model based on Korean. The processor 210 can further learn one or more second comments into the pre-learning model 510. In other words, the processor 210 can fine-tune the pre-learning model 510 using one or more second comments. As a result, the learning model can be constructed in a way that outputs the features of the input second comments as vectors. In one embodiment, the processor 210 can also determine recommended comments based on the pre-learning model 510, which has been fine-tuned by other devices or processors.

[0051] However, the language used to fill in the first or second comment is not limited to Korean; it can be filled in various languages ​​depending on the user's nationality or place of residence. The processor 210 can utilize a pre-learning model 510 suitable for the language used to fill in the first comment to train one or more second comments to be filled in the same language as the first comment.

[0052] Figure 6 is a diagram illustrating one example of generating the first vector in step S420 of Figure 4. In various embodiments, memory 220 may store commands for execution in processor 210, which may perform the following actions by executing the commands. Processor 210 may be configured to input first information into a learning model to generate the first vector.

[0053] In various embodiments, the processor 210 may further include a preprocessing unit 610 that preprocesses the first information to make it suitable for input to the learning model 620. In one embodiment, the preprocessing unit 610 can divide the input text information into words to generate tokenized preprocessed information. For example, the first comment of the first information may include more than one word, and the preprocessing unit 610 can tokenize each of the more than one word of the first comment to generate the first preprocessed information.

[0054] The processor 210 can be configured to input first preprocessed information into the learning model 620 to generate a first vector. The first vector may include feature information extracted from the first comment. The processor 210 can store the first vector in memory 220 in association with the first comment.

[0055] Figure 7 is a sequence diagram illustrating the determination of one example of a recommended review in step S330 of Figure 3. In step S710, the processor 210 may obtain one or more second vectors corresponding to each of the more than one second review. The second vector may include feature information extracted from the second review.

[0056] In one embodiment, the processor 210 may input one or more second comments into the learning model 620 to obtain one or more second vectors. In one embodiment, the processor 210 or an external device may generate one or more second vectors corresponding to each of the one or more second comments and store them in memory 220. The processor 210 may obtain one or more second vectors from memory 220. In one embodiment, the one or more second comments may be distributed and stored in one or more external server devices. The processor 210 may transmit a query request to one or more external server devices to retrieve one or more second comments via communication circuit 230. In response to the query request, one or more external server devices may transmit one or more second comments to the processor 210.

[0057] In step S720, the processor 210 may determine the similarity between the first vector and one or more second vectors. In various embodiments, the processor 210 may determine the similarity based on the absolute value of the distance between the first vector and one or more second vectors. For example, the processor 210 may determine the similarity based on the absolute value of the distance between the first vector and one or more second vectors. In other words, the similarity between the first vector and the second vector may refer to the distance between the first vector and the second vector.

[0058] In step S730, the processor 210 may determine one or more vectors in the second vector as similar vectors. In one embodiment, the processor 210 may use the K-Nearest Neighbors algorithm to determine a fixed number (e.g., K) of vectors in one or more second vectors that are similar to the first vector as similar vectors. In one embodiment, the processor 210 may determine vectors in one or more second vectors that have a similarity of at least a first threshold value to the first vector as similar vectors based on their similarity scores. In one embodiment, the processor 210 may determine a fixed number of vectors in one or more second vectors as similar vectors based on their similarity scores, in descending order of similarity.

[0059] In step S740, the processor 210 can determine the comments corresponding to each of the more than one similarity vectors in the more than one second comment as more than one recommended comment.

[0060] FIG8 is a sequence diagram illustrating the operation method of the apparatus 120 according to various embodiments of the present invention. In various embodiments, the memory 220 may store commands to be executed in the processor 210 and information processed in the processor 210, which may perform the following operation method by executing commands.

[0061] In step S810, the processor 210 may further obtain from the terminal a request for additional comments other than one recommended comment. For example, in addition to the recommended comments displayed by the terminal's display device, the user may further request one or more comments from the second comments that are similar to the first comment.

[0062] In step S820, the processor 210 may, in response to an additional comment request, identify one or more comments that are not among the more than one recommended comments in the more than one second comment as additional comments.

[0063] In various embodiments, the processor 210 may determine one or more second comments as additional comments based on the similarity between the first vector and one or more second vectors. In one embodiment, the processor 210 may determine vectors among one or more second vectors that have a similarity of at least a second threshold value as additional vectors based on the similarity. The second threshold value may be less than the first threshold value. For example, an additional vector may be one whose similarity to the first vector does not reach the first threshold value but is at least the second threshold value. In one embodiment, the processor 210 may determine a fixed number of vectors among one or more second vectors corresponding to one or more second comments that are not recommended comments as additional vectors, in descending order of similarity. The processor 210 may determine the comments corresponding to the additional vectors as additional comments.

[0064] In step S830, the processor 210 can transmit additional comment information of one or more additional comments to the user's terminal 110. The additional comments of the additional comment information can be displayed to the user through the display device of the user's terminal 110.

[0065] FIG9 is a sequence diagram illustrating the operation method of the apparatus 120 according to various embodiments of the present invention. In various embodiments, the memory 220 may store commands to be executed in the processor 210 and information processed in the processor 210, which may perform the following operation method by executing commands.

[0066] In step S910, the processor 210 can obtain consent / disagreement information from the user's terminal 110, which indicates whether the user agrees with more than one recommended reviewer. It can determine whether the user agrees (or recommends) or disagrees (or does not recommend) at least one of the more than one recommended reviewers displayed on the terminal's display device, thereby generating consent / disagreement information via the display device (or input device). The processor 210 can obtain the consent / disagreement information generated by the user's terminal 110 via the communication circuit 230.

[0067] In step S920, the processor 210 may associate the user's consent or disagreement with the comments of one or more recommended commenters belonging to one or more second comments, based on the consent or disagreement information. The processor 210 may store the information associated with whether or not the user agrees with one or more recommended commenters in memory 220.

[0068] In step S930, in various embodiments, the processor 210 may determine one or more comments that do not belong to one or more recommended comments as additional comments based on whether the user agrees with one or more recommended comments.

[0069] In various embodiments, processor 210 may, in response to a user's agreement to an association result of a recommended comment, identify one or more comments similar to a recommended comment as additional comments. In one embodiment, processor 210 may determine the similarity between a vector corresponding to a recommended comment and a second vector corresponding to one or more second vectors that do not belong to one or more recommended comments, and determine one or more additional vectors based on the similarity.

[0070] In various embodiments, the processor 210 may determine the similarity based on the absolute value of the distance between the vector corresponding to a recommended review and one or more second vectors. In one embodiment, the processor 210 may, based on the similarity, determine one or more vectors among the one or more second vectors that do not belong to the recommended review and have a similarity of at least a third threshold as additional vectors. In one embodiment, the processor 210 may, based on the similarity, determine a fixed number of vectors among the one or more second vectors that do not belong to the recommended review and are in descending order of similarity as additional vectors.

[0071] The processor 210 can determine a comment corresponding to one or more additional vectors as one or more additional comments.

[0072] In step S940, the processor 210 can transmit additional comment information of one or more additional comments to the user's terminal 110. The additional comments of the additional comment information can be displayed to the user through the display device of the user's terminal 110.

[0073] In the flowcharts of this invention, the steps of the method or algorithm are described sequentially. However, in addition to performing the steps sequentially, they may also be performed in an order that can be arbitrarily combined according to the invention. The description of the flowcharts of this invention does not preclude changes or modifications to the method or algorithm, and does not imply that any step is necessary or preferred. In one embodiment, at least some steps may be performed in parallel, repeatedly, or heuristically. In another embodiment, at least some steps may be omitted, or other steps may be added.

[0074] Various embodiments of the present invention can be implemented in the form of software on a machine-readable storage medium. The software can be software used to implement the various embodiments described in the present invention. Programmers in the art to which this invention pertains can deduce the software based on the various embodiments described in the present invention. For example, the software can be a program that includes machine-readable commands (e.g., code or code segments). The machine is a device capable of operating according to commands invoked from the storage medium, such as a computer. In one embodiment, the machine can be device 120 of the various embodiments described in the present invention (see FIG2). In one embodiment, the processor 210 of the machine executes the invoked command, thereby enabling the components of the machine to perform functions corresponding to the command. In one embodiment, the processor 210 can be processor 210 (210, see FIG2) of the embodiments described in the present invention. The storage medium can refer to all kinds of recording media capable of being read by a machine and storing data. Storage media may include, for example, ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc. In one embodiment, the storage media may be memory 220 (220, see Figure 2). In one embodiment, the storage media may also be implemented in the form of being distributed across computer systems connected to a network. In this case, software can be distributed across computer systems for execution. In another embodiment, the storage media may be non-transitory storage media. Non-transitory storage media refers to tangible medium that is not related to the semi-permanent or temporary storage of data, excluding transiently transmitted signals.

[0075] The technical concept of the present invention has been described above according to various embodiments, but the technical concept of the present invention includes various substitutions, variations, and modifications that can be understood by those skilled in the art to which the present invention pertains. Furthermore, it should be understood that such substitutions, variations, and modifications may be included within the scope of the appended patent applications. [Simplified Explanation of the Diagram]

[0023] Figure 1 is a diagram illustrating a system according to various embodiments of the present invention. Figure 2 is a diagram illustrating an example of an apparatus according to an embodiment of the present invention. Figure 3 is a sequence diagram illustrating an example of an operation method of an apparatus according to various embodiments of the present invention. Figure 4 is a sequence diagram illustrating an example of generating a first vector in step S320 of Figure 3. Figure 5 is a diagram illustrating an example of training a learning model by a second review in step S410 of Figure 4. Figure 6 is a diagram illustrating an example of generating a first vector in step S420 of Figure 4. Figure 7 is a sequence diagram illustrating an example of determining a recommended review in step S330 of Figure 3. Figure 8 is a sequence diagram illustrating an example of an operation method of an apparatus according to various embodiments of the present invention. Figure 9 is a sequence diagram illustrating an example of an operation method of an apparatus according to various embodiments of the present invention.

Claims

1. An apparatus for determining product reviews, comprising: The device comprises one or more processors and one or more memory stores commands for execution by the processors. When executing the commands, the processors perform the following: Obtaining first information from a user's terminal regarding a first review of a product; Generating a first vector corresponding to the first review based on the first information; Determining one or more recommended reviews similar to the first review from among one or more second reviews of the product based on the first vector; Transmitting the recommendation review information of the one or more recommended reviews to the terminal; Obtaining from the terminal additional review requests for reviews other than the one or more recommended reviews; In response to the additional review requests, determining one or more additional reviews from among the reviews not belonging to the one or more recommended reviews; and transmitting the additional review information of the one or more additional reviews to the terminal.

2. The apparatus of claim 1 is configured such that, when generating the first vector, one or more processors input the first information into a learning model to generate the first vector as the output of the learning model, the learning model being a learner that converts product reviews into vectors.

3. The apparatus of claim 2, wherein the learning model is a model that learns by using one or more second reviews written for the product as learning data.

4. The apparatus of claim 2, wherein the learning model is a BERT model.

5. The apparatus of claim 1, configured such that, when determining the aforementioned one or more recommended reviews, the aforementioned one or more processors perform the following: obtaining one or more second vectors corresponding to each of the aforementioned one or more second reviews; determining the similarity between the aforementioned first vector and each of the aforementioned one or more second vectors; determining one or more similar vectors among the aforementioned one or more second vectors based on the aforementioned similarity; and determining the review among the aforementioned one or more second reviews corresponding to each of the aforementioned one or more similar vectors as the aforementioned one or more recommended reviews.

6. The apparatus of claim 5 is configured such that, when determining the similarity between the first vector and one or more second vectors, the processor determines the absolute value of the distance between the first vector and one or more second vectors as the similarity.

7. The apparatus of claim 5 is configured such that the aforementioned one or more processors store the aforementioned first comment and the aforementioned first vector in the aforementioned one or more memory.

8. The apparatus of claim 6 is configured such that, when determining one or more similar vectors, the processor determines the vector among the one or more second vectors whose similarity to the first vector is at or above a first threshold value as one or more similar vectors.

9. The apparatus of claim 8, configured such that, when determining one or more additional comments, the processors perform the following: Based on the aforementioned similarities, determine one or more additional vectors from the aforementioned one or more second vectors, wherein the similarity between the additional vectors and the first vectors is less than the aforementioned first threshold and is greater than or equal to the second threshold; determine the comments corresponding to each of the aforementioned one or more additional vectors as the aforementioned one or more additional comments; wherein the aforementioned second threshold is less than the aforementioned first threshold.

10. The apparatus of claim 5 is configured such that the aforementioned one or more processors are configured to: obtain consent or dissent information from the aforementioned terminal, the consent or dissent information indicating whether the aforementioned user agrees with at least one commenter among the aforementioned one or more recommended comments; and, based on whether the aforementioned user agrees with the aforementioned at least one comment, identify one or more comments among the aforementioned one or more second comments that do not belong to the aforementioned one or more recommended comments as the aforementioned one or more additional comments.

11. A method for determining recommended reviews of a product, implemented in an apparatus comprising one or more processors and one or more memories storing commands to be executed by the one or more processors, the method comprising the steps of the one or more processors as follows: obtaining first information of a first review of a product by a user from a user's terminal; generating a first vector corresponding to the first review based on the first information; determining one or more recommended reviews similar to the first review from one or more second reviews of the product based on the first vector; and transmitting the recommended review information of the one or more recommended reviews to the terminal; further obtaining from the terminal a request for additional reviews other than the one or more recommended reviews; in response to the additional review request, determining one or more additional reviews from reviews not belonging to the one or more recommended reviews; and transmitting the additional review information of the one or more additional reviews to the terminal.

12. The method of claim 11, wherein the step of generating the first vector by one or more processors includes the following steps: inputting the first information into a learning model to generate the first vector as the output of the learning model, the learning model being a learner that converts product reviews into vectors.

13. The method of request item 12, wherein the learning model described above is a model learned using one or more second reviews filled out for the product described above as learning data.

14. As in request item 13, wherein the learning model described above is the BERT model.

15. The method of claim 11, wherein the step of the processor determining the one or more recommended reviews comprises the following steps: obtaining one or more second vectors corresponding to each of the one or more second reviews; determining the similarity between the first vector and each of the one or more second vectors; determining one or more similar vectors among the one or more second vectors based on the similarity; and determining the review corresponding to each of the one or more similar vectors among the one or more second reviews as the one or more recommended reviews.

16. The method of claim 15, wherein the step of the processor determining the similarity of the first vector and the second vector comprises the step of: determining the absolute value of the distance between the first vector and the second vector as the similarity.

17. The method of claim 16, wherein the step of the processor determining the one or more similar vectors comprises the following steps: determining the vectors among the one or more second vectors whose similarity to the first vector is above a critical value as the one or more similar vectors.

18. A computer-readable recording medium recording commands that, when executed by one or more processors, cause the one or more processors to perform actions, wherein the commands are configured such that the one or more processors: obtain first information from a user's terminal regarding a first review of a product; generate a first vector corresponding to the first review based on the first information; determine, based on the first vector, one or more recommended reviews among one or more second reviews of the product that are similar to the first review; transmit the recommendation review information of the one or more recommended reviews to the terminal; further obtain from the terminal a request for additional reviews other than the one or more recommended reviews; in response to the additional review request, determine one or more additional reviews from reviews that are not among the one or more recommended reviews; and transmit the additional review information of the one or more additional reviews to the terminal.

19. The computer-readable recording medium of claim 18 is configured such that, when generating the first vector, one or more processors input the first information to a learning model to generate the first vector as the output of the learning model, the learning model being a learner that converts product reviews into vectors.

Images