KR20190088131A - Deep learning module and curation system therewith for customized travel goods recommendation - Google Patents

Abstract

Disclosed are a travel package recommendation deep learning module and a personalized travel package curation system using the travel package recommendation deep learning module. According to the present invention, the travel package recommendation deep learning module is trained by using attribute data of a travel package clicked by a user who accessed on a certain Internet protocol (IP) address as a feature data and using the attribute data of a travel package booked by the user after the user clicked travel packages as label data. According to the present invention, consumer satisfaction increases, booking rate for packages increases, and canceling rate decreases.

Description

{DEEP LEARNING MODULE AND CURATION SYSTEM THEREWITH FOR CUSTOMIZED TRAVEL GOODS RECOMMENDATION} <br> <br> <br> Patents - stay tuned to the technology DESIGN LEARNING MODULE AND CURATION SYSTEM THEREWITH FOR CUSTOMIZED TRAVEL GOODS RECOMMENDATION

The present invention relates to a travel product recommending deep learning module and a personal travel product curation system using the travel product recommend deep learning module. More particularly, the present invention relates to a travel product recommendation deep learning module, And a personal travel product curation system using the deep-learning module and the deep-learning module which is recommended for the travel product.

Travel is becoming more individualized and diversified. It is diversified and individualized in past package trip, free travel or local travel. As a result, the detailed composition of travel products provided by travel agencies has diversified, and the number of travel goods has increased to tens of thousands (40,000 to 50,000).

On the other hand, there is a tendency for travelers to search for and purchase online travel products online without the help of travel agencies. Selecting a travel product by searching for it online can be a good way to find out what kind of trip you are going to take on a package tour, a free trip or a local trip, what price range of trips to choose, where to go, Tour, and so on, it is not only difficult to make the right choice among the many travel products, but also takes too much time to choose.

In the case of a travel agency that sells travel products, if a consumer chooses a travel product but the consumer does not make the proper choice, the reservation rate is lowered and the cancellation rate is increased. As a result, Satisfaction is also lowered.

Therefore, there is a great need for a recommendation service that easily and conveniently finds the best travel products among the many travel products that are sold online.

According to this necessity, a tour product search system based on natural language processing and a method thereof (Korean Patent Laid-Open No. 10-2008-0058569), a travel product search system using the Internet, and a travel product recommendation method using the same 10-2000-0031026) have been proposed.

However, online travel product recommendation is based on the fact that the user is likely to use only one or two times a year on average, so that the probability of frequent use is low, that the user is not used to membership, , It is not easy to appropriately perform the method disclosed in these patent publications. Particularly, the recommendation methods disclosed in the above-mentioned patent publications disclose a method in which a consumer who is a member of a recommendation service server searches for and recommends a travel product based on the conditional word input, And there is also a problem in that it is difficult to select and input an appropriate condition search word because the detailed knowledge of the travel product is insufficient.

Accordingly, in order to effectively provide a traveler with a recommendation service for a travel product online, when a traveler accesses a travel product search engine regardless of whether or not a member is a member and searches for a travel product in a free manner or clicks on a presented travel product, This process is tracked in real time to individually extract the product selection pattern of the traveler, to predict the desired travel product or reservationable travel product, to continuously recommend the travel product predicted in real time during the travel product search to the traveler You need a system.

In addition to the recommendation methods disclosed in the above patent publications, statistical recommendation methods, recommendation methods based on product characteristics, recommendation methods based on collaborative filtering, and the like have been used as recommendation methods for travel products on the Internet.

The statistical recommendation method is a method of statistically analyzing popular products and recommending them, and it is difficult to reflect personal needs, tastes and plans of travel applicants, and has a disadvantage of recommending irrelevant to personal characteristics of travel applicants.

The recommendation method based on the product characteristic is a recommendation method of recommending another recreational product to the customer who viewed the recreational product. In order to use such a method, the travel product is reflected to the personal desire, taste and plan, However, it is difficult to classify the travel products according to the characteristics of the goods, and the classification standard is also unclear.

The collaborative filtering based recommendation method is a method of recommending travel products that similar users refer to or recommending highly related travel products. However, the collaborative filtering based recommendation method has a disadvantage in that it is difficult to recommend highly accurate travel products that exactly fit consumer desires in that the similar users can make completely different choices according to the detailed attributes of travel goods.

Korean Unexamined Patent Application Publication No. 10-2008-0058569 (June 26, 2008) Korean Unexamined Patent Application Publication No. 10-2001-0110523 (December 13, 2001)

The present invention has been made to solve the disadvantages of the above-described conventional online travel product recommendation service, and a first problem to be solved by the present invention is to provide an online travel product recommendation service in which each individual consumer of a travel product accurately analyzes the desire for a current travel product, And a travel commodity curtailing system using a deep running module and a travel commodity deep running module that can recommend an appropriate travel commodity.

A second problem to be solved by the present invention is to provide a travel product recommending deep learning module capable of recommending a responsive travel product based on a real-time search pattern of a travel product of an individual consumer and a personalized travel product recommendation module And to provide a system of analysis.

A third problem to be solved by the present invention is to provide a travel product recommending deep learning module capable of recommending a detailed and precise travel product utilizing up to detailed attributes of a travel product and a personal travel product curation system using the travel product recommending deep learning module .

The above-described objects of the present invention are accomplished by using attribute data of a travel article clicked by a user connected at a certain IP address (Internet Protocol Address) as the feature data, It is possible to solve this problem by using the travel product recommending deep learning module which has learned the attribute data of the travel product as label data.

The above objects of the present invention can also be accomplished by providing a system and method for realizing a travel product recommending deep learning module that, when a real-time user accesses and searches for a travel product, stores log data of the real- A travel product search engine for receiving a travel product deduced as a reservationable product in the travel product recommending deep learning module and displaying the travel product as a recommended product on a screen of a travel product clicked by a real time user, And a log analyzing module for extracting the travel product attribute data included in the log data, processing it into a form that can be used as input data in the travel product recommending deep learning module, and providing it to the travel product recommending deep learning module Product recommendation Using the Deep Learning Module for individual travel product curation system And it can be solved.

According to the present invention, when a traveler accesses a travel goods search engine regardless of whether or not a member is a member and searches for a travel product in a free manner or clicks on a presented travel product, this process is tracked in real time, It is possible to predict a desired travel product or a reservationable travel product and to continuously recommend the travel product predicted in real time during the search of the travel product to the traveler so that the customer satisfaction is increased and the product Not only the reservation rate is increased but also the cancellation rate is decreased and the productivity and the profit rate of the travel agency can be improved.

FIG. 1 shows an example of log data used in a travel product recommending deep learning module according to the present invention.
FIG. 2 is a table showing trip property data extracted from the log data shown in FIG.
FIG. 3 shows an example in which a travel product recommending deep learning module according to the present invention is composed of RNN (Recurrent Neural Networks).
FIG. 4 is a configuration diagram of a personalized travel product curation system using a travel product recommending deep learning module according to the present invention.
FIG. 5 is an exemplary screen for recommending a travel product using a tour product curation system according to the present invention.

Hereinafter, with reference to the accompanying drawings, a detailed description will be given of a tour product recommending deep learning module according to the present invention and a personal tour product curation system using the travel product recommending deep learning module.

As shown in Figs. 1 and 4, when a user accesses a travel agent homepage (e.g., http://www.hanatour.com/) or a travel portal homepage (e.g., http: //www.laontrIP.com) When searching for a product or clicking on a specific travel product, a travel product search engine executed on these homepage servers automatically generates user log data and stores the log data file in the storage as a log data file.

As shown in Fig. 2, in the user log data, the IP address of the connected user, the clicked date and time (date and time of inquiry), the clicked travel agent code (inquired travel agent code) A variety of travel goods subcode (the subcode of the travel item clicked), the attributes of the clicked travel goods (departure date, number of clicks, area name, country name, city name, airline, travel type, price, Information can be extracted and organized. In the user log data, the IP address of the user who reserved the trip product, the date and time of the reservation, the booked travel agent code, the booked travel product code, the booked travel product subcode, , Region name, country name, city name, airline, type of travel, price, travel period).

4, the travel goods D / B 5 is provided in the travel goods search engine 2 and the attributes of the travel goods provided by the travel agency or the like in the travel goods D / B 5, (Package, free), area name, country name, city name, airline, hotel, travel period, schedule, meal, vehicle, other options, etc.) are displayed and displayed on the user's search screen.

As shown in FIG. 3, the travel product recommending deep learning module according to the present invention extracts attribute data of a travel product clicked by a user connected from a certain IP address from the log data extracted from the log data, feature data, and the attribute data of the tour product reserved after the user clicks on the travel products is used as label data. Attribute data of travel products clicked (queried) in the process of searching for a travel product before a certain user reserves a travel product are sequence data or time series data that are temporally sequential, The number of travel products to be clicked is different before the reservation of the product is reached. Therefore, as shown in FIG. 3, the travel product recommending deep learning module according to the present invention includes a recurrent neural network (RNN) capable of learning by inputting sequence data or time series data, It is desirable to use an LSTM (Long Short Term Memory). Also, since the number of travel products to be clicked is different for each user before reaching the reservation of the travel product, it is preferable to use the Many to Many model as shown in FIG. At this time, using the property data of the tour product that has been continuously clicked and the property data of the booked travel product, state function and an output function (label function) of a weight value and a bias value.

As shown in FIG. 4, the personal travel product curation system according to the present invention includes: the learned travel recommendation deep learning module 4; Engine 2 and a log analysis module 3 for preprocessing the travel goods search engine log data of the user so as to be processed in real time by the travel product recommending deep learning module 4.

When the real-time user accesses the travel product search engine 2, the travel product search engine 2 stores the log data of the real-time user and transmits the log data to the log analysis module 3, 4) receives the tour product deduced as the reservationable product and displays it as a recommended product on the travel product screen clicked by the real-time user.

The log analysis module 3 receives the real-time user log data from the travel product search engine 2, extracts the travel product attribute data included in the log data, and inputs it to the travel product recommending deep-learning module 4 Data to be used as data and provides it to the travel product deep learning module 4.

The travel product deep learning module 4 continuously receives attribute data of the travel goods clicked (inquired) by the real-time user, deduces the property data of the travel goods that can be reserved at each click step, (3) to the travel product search engine (2), and the travel product search engine (2) displays the search result on the inquiry screen of the real time user (1).

Referring to FIGS. 4 and 5, an exemplary travel service recommendation service using the individual travel goods curation system according to the present invention will now be described.

4, the travel product D / B 5 is provided in the travel product search engine 2 and the attributes of the travel product (departure date, price, type (package, Data is stored together with the travel commodity code so that the user can be prompted to search the user's query or click on the travel commodity (click) Lt; / RTI &gt; The travel product search engine 2 provides the current travel product data to the log analysis module 3 and the log analysis module 3 provides the travel product recommendation deep learning module 4 300).

Before the real-time travel product recommendation service, the training of the travel recommendation deep-learning module should be performed first. To this end, the travel product search engine 2 provides an existing user log data file to the log analysis module (201). The log analysis module 3 provided with the user log data file extracts attribute data of the inquiry items (click items) before the reservation per user (or IP address) and the actual reservation items of the user from the user log data, The data is processed into data of a form that can be learned by the product recommending deep learning module 5 (301) and provided to the travel product recommending deep learning module 5. The log analysis module 3 classifies, analyzes and organizes the log data of a large number of users included in the user log data file on the basis of the IP address, processes the log data, and provides it to the travel product recommending deep learning module 5. The travel product recommending deep learning module 5 uses the attribute data of the travel goods that a certain user clicked based on the Internet Protocol Address as a feature data, (400) using attribute data of the travel goods reserved as label data. The learning is repeatedly performed using the attribute data of a large number of users in units of IP addresses. 3, when the RNN (LSTM) is used, the change of travel product selection by the user at every click is inputted into the cell of the next step as a state function, and then, The learning product recommending deep learning module 5 learns the trip product in the process of receiving the travel product clicked by the real time user 1 in the future, The real-time user 1 can accurately deduce the attribute data of the desired travel product.

The learned travel product recommending deep learning module 5 is usefully used in a personal tour product curation system so that the real-time user 1 can quickly and accurately recommend a travel product.

As shown in FIG. 4, when the real-time user 1 accesses the travel goods search engine 2 and searches for a travel product, the property data of the travel product clicked (inquired) by the real- , And transmits it to the log analysis module 3 in real time, the log analysis module processes the attribute data of the travel goods clicked (inquired) into data of a form that can be processed by the travel goods recommending deep learning module 4 And transmits it to the travel product recommending deep learning module 4 (304). The trip product deep running module 4 deduces (401) the property data of the travel product that can be reserved by the real time user using the travel product property data clicked by the real time user and the change, (402), and transmits the travel commodity code to the log analysis module (3) (403). The log analysis module 3 transfers the reservationable travel product code to the travel product search engine 2 and the travel product search engine 2 searches the travel product search screen of the real- As a recommended product (204).

The travel product search engine 2 preferably transmits the log data to the log analysis module when the number of travel goods clicked by the real-time user reaches a certain number or more, or when the connection time exceeds a predetermined time. By doing this, more real-time users can more accurately infer the desired travel products and recommend them.

According to the present invention, detailed recommendation is made through detailed attribute analysis of the travel goods. FIG. 5 shows an embodiment in which a recommended product is displayed on the travel product inquiry screen of the real-time user 1 through this process.

As described above, according to the present invention, the correlation between the inquired trip product and the booked travel product is machine-learned, the travel product recommending deep learning module 4, and the log analysis module 3 for analyzing travel product click data by the real- When a user clicks on a specific travel product after searching for a travel product on an online travel product site, the log analysis engine analyzes and processes the property data of the travel product viewed by the real-time user, The Deep Learning module recommends the real-time user by deducing the real-time user's most desired tour product.

1: Real-time user terminal
2: Travel goods search engine
3: Log analysis module
4: Deep Learning Module
5: Travel goods D / B

Claims (4)

As a deep running module,
The attribute data of the travel product that the user who has accessed from the Internet Protocol Address (IP Address) clicked on is used as the feature data and the attribute data of the travel product that the user has reserved after clicking on the travel products, (label data).
The method according to claim 1,
Wherein the deep learning module is a Recurrent Neural Network (RNN) that uses attribute data of a travel article clicked by a user as input sequence data.
The travel goods recommending deep learning module according to claim 1 or 2;
When the real-time user accesses and searches for a travel product, the log data of the real-time user is stored and transmitted to a log analyzing module (hereinafter referred to as a log analyzing module), and the tour product deduced as a reservation- A travel product search engine that displays a recommended product on a travel product screen clicked by a real time user;
The log data of the real time user is received from the travel goods search engine, the travel product attribute data included in the log data is extracted, and the travel product recommendation deep learning module processes the travel product recommend deep running And a log analysis module provided to the module.
The method of claim 3,
Wherein the travel goods search engine transmits the log data to the log analysis module when the number of travel goods clicked by the real time user reaches a certain number or more, Individual travel curation system using learning module.

Images