Semantic Framework for Facilitating Product Discovery

Abstract

With the rapid growth in the field of e-commerce, online shopping has become a major part in lives of people. Online stores provide search engines for product discovery. Traditional search engines are working on syntactic approach, because of which these search engines are not able to extract precise result. Since products catalogues comes from diferent sources, there comes need of binding the information of these product catalogues into a single taxonomy for better categorization of products. By using semantic web technologies, we can represent all the information in machine understandable form. Semantic search engine use knowledgebase to store and retrieve the information. Knowledgebase uses a semantic model that define all the relationships, classes and properties. Instances are annotated according to this model and a knowledge graph is created. This intelligent binding of data helps the semantic search engine to understands the user query and provides accurate result. We propose a semantic search engine framework for facilitating product discovery which uses an ontology for product categorization and annotate all the product catalogues from diferent retailers into a single knowledge graph or RDF. After that the RDF store is used for storing and retrieving data using semantic queries.

Keywords

• eol>E-Commerce
• Knowledge Graph
• Ontology
• Product Discovery
• Semantic Search
• RDF

1. Introduction

Online shopping is comparatively easier and better than traditional shopping because nowadays everything is available to us at our door steps just what we need to have is a good availability of internet. An ineficient product searching has been a major cause of decrements in sales, continuous poor service [ 1,3,8 ] can afect the business likely for evoking new customers as well. As stated by [ 15 ] Statista Statistic report in 2018, an estimated 1.8 billion Figure 1: Shoppers’Frustration with Retail Site Search. people around the world buy goods online. During the same year, global e-retail sales amounted to 2.8 trillion U.S. dollars and projections show a growth of up to 4.8 trillion U.S. dollars by 20. requirements [ 4,20 ]. There are certain problems with the

The Fig. 1 [ 13 ] shows the snags which online cus- traditional approaches. Interoperability is also a major tomers face when they search for products on e-Commerce problem that arises due to the heterogeneous formats of websites. As per the statistics[ 14 ] published on Market- information representation[ 11 ]. Because of innovation ing Charts.com in June 2018, Data Source: RichRelevance, new features are added to products frequently which remany customers end up with irrelevant product results quires continuous changes in schema. The basic problem (28 %), unable to find the product they need (24 %) , search in keyword-based search is it only finds keywords of function does not recognize the words used by particular user query on the web and list all the pages in result[ 18 ]. customer(18 %), unclear search box(10 %) and many more. For example- if you are searching on a keyword-based The absence of HTML standard to introduce the heteroge- search engine "Books about hotel", this will display you neous e-commerce information which results in the low the web pages about “hotels”, “books”, “hotel booking” quality of web crawler results, which makes clients to in- and “books about hotels” but actually you were searching vest time and energy for arranging and comparing items for the books written about hotels. The keyword-based between sites and picking the correct item that suits their search gives better hit ratio because it works on two keywords hotel and books but semantic precision and recall is low in this. We want it to understand the intent of user but it only understands the keyword from the web pages.

Web technologies are changing rapidly from static web to progressive web, then to semantic web [ 23 ]. Semantic web helps machine to interpret the meaning of data. Semantic web technologies can be applied in various fields ACI’21: Workshop on Advances in Computational Intelligence at ISIC 2021, February 25-27, 2021, Delhi, India " [e-mail] (P. Bhutani); [e-mail] (S. K. Baranwal); [e-mail] (S. Jain) 0000-0002-7432-8506 (S. Jain)

© 2021 Copyright for this paper by its authors. Use permitted under Creative CPWrEooUrckReshdoinpgs IhStpN:/c1e6u1r3-w-0s.o7r3g CCoEmmUoRns LWiceonsrekAstthribouptionP4r.0oIncteerenadtiionnagl s(CC(CBYE4U.0)R.-WS.org) facilitating product discovery.

2. Related Work

There are two ways for product categorization one is using machine learning techniques and another one is using lexicon-based techniques[ 2 ]. As our work is on semantic web technologies so we are discussing here work related to semantic web technologies. The authors in [ 5 ] have described about product information retrieval framework that uses OA-VSM and SPARQL-based approach rather Figure 2: Result on Keyword-based Search Engine. than VSM-based keyword search. Ontology based adaption of vector space model retrieval method helps the user to get better result. The authors in [ 6 ] have introduced the importance of availability of machine understandable where data integration improves the quality like oil and information. Machine understandable information has gas industry, pharma industry, movies/series searching. the potential to impact many significant web applications In all these fields a huge amount of data is generated that includes searching. Two mechanisms are discussed every day. Semantic search engine has its own benefits that have increasing search results from the semantic web. which results comes in relevant matches rather than pro- Angelo A. Salatino et al. [7] presented CSO classifier as viding unnecessary product information. As everything a tool which classifies text according to the Computer is defined using relationship it’s easy to understand cus- Science Ontology. CSO Classifier takes input in form of tomer’s requirements which results in better relationship unstructured text, title, and keywords of a research paper of enterprise-customer. If product sites documents are and classify the research area of paper as output. E. Peis designed semantically, one may get better results while et al [ 9 ] states that recommender systems based on sesearching and surely that will profitable to the business. mantic technologies give out high performance because

For online business product database is the wealth of they are based on a knowledge base mostly defined by e-business and the site is its appearance, the "virtual" an ontology. Using ontology solve interoperability probexhibit for clients. For obtaining this we have used the lem, improve performance in social networks, represent knowledge base to store the data efectively. Knowledge information semantically and improve searching result. base work like human brain to organize information in The authors in [ 16 ] stated the importance semantic an abstract way. In knowledge-base a semantic model web mining which means that semantic adds meaning to is present to better classify the data using classes, sub- the data but web mining use to filter out that data more classes, relationships and instances. Ontology has the accurately. The basic idea is to extract the useful inforability to define such model. The goal is to create a seman- mation from the bulk and for that we need standardized tic model for sharing information about a specific domain. ontologies. Example- Goodrelations, e-class OWL. Kerie There are various product classification standards which A. [ 17 ] stated the current limitation of searching on the exists. These help for better online data handling, dealing web and the reason of these problems are syntactic and with business related terms and services. Few of them semi semantic approach. On the basis of ontology, they are eClassOWL , GoodRelations , Global Product Clas- have tried to correct these problems. Here they have sification, Schema.org etc. We have also developed an made EEPS (Ethiopian Export Products and Services Doontology for categorization of electronic devices. The main) ontology which is consistent, flexible and moderate next step data integration involves knowledge graph that in size which makes the management of the information represents the information in RDF is used for data in- better. They study [ 19 ] the challenges that traditional tegration. Benefit of using knowledge graph is that it e-commerce faces and overview of how semantic web provides expressivity, performance and standardization. technologies overcome these challenges in e-commerce A triple store or RDF store is used for the storage and ifeld. They have developed a semantic search engine retrieval of information from RDF using SPARQL queries for e-commerce field using GOODRELATION ontologies. with the help of SPARQL endpoint.The key idea here The authors in [ 21 ] designed a framework based on onis to semantically annotate all the products catalogues tology for the purpose of search and retrieval of products. according to a single taxonomy and store it into triple They find out that the ontology created for searching perstore. Our contribution is mainly in two directions i.e., forms better for searching purpose rather than catalog we have created an ontology for product categorization. side ontologies. Necula S. [ 22 ] in their study find out that and a working prototype using semantic technologies for customer consider knowledge graph is important for an e-commerce website. Another finding of their study is that searching a product by categories and subcategories gives better result.

3. Proposed Framework

To overcome the problem of irrelevant search of products

on online store our work uses a semantic search engine. The aim of our work is to provide customer best product according to their requirements. Figure 3 presents a general architecture of our proposed framework. The proposed system has two interfaces, one for retailers in order to store new products and another one for customers to get recommendation according to their input specification about products. We have divided proposed framework into two modules. In first module i.e., in knowledge modeling we will create schema for knowledgebase and after that instances are populated. All these data are then stored in triple store. After that in second module i.e., in knowledge consumption model end user enters the specification about products and a SPARQL query is processed over knowledge base and desired results are fetched according to the end users’ requirements.

4. Materials and Methods

In this section we discuss about various tools and technologies that can be used to implement the proposed framework

4.1. Existing Product Classification

Standards

There are various product classification standards which exists. Out of these some of them are mentioned below which are useful for us. • eClassOWL: eClassOWL released in 2004. eClassOWL is intended to be utilized for product classification. It has 30,000 classes and 5,000 properties of product features. • GoodRelation: GoodRelations is a vocabulary used for sharing e-commerce information. It has 30,000 classes and 15 properties of product features. Now it is integrated into schema.org and supported by Google, Yahoo etc. • Global Product Classification: GPC is utilized for grouping all the products into a common language. The building block of GPC is an item code known as a brick. There are bricks for everything from a vehicle to a bottle of milk. The most elevated level of the characterization is a fragment, which is characterized as a specific industry. For instance, a bottle of milk has a place with the food, beverages and tobacco section. • UNSPSC: UNSPSC is a vocabulary about products and services used in e-commerce field. It has 20,792 classes. UNSPSC empowers expenditure analysis at gathering levels pertinent to your requirements. For our proposed framework we have created our own product classification ontology.

For our proposed framework we have created our own product classification ontology.

4.2. Triple Store

This list shows notable triple store, APIs, and other stor

ages that have implemented the W3C SPARQL standard. • Apache Jena: Jena supports all operating system like Linux, Windows etc., with a Java Virtual Machine. It is an open-source software. It gives us an API to take out data from and write to RDF file. Apache Jena is completely FREE as open-sourced Java-based software from Apache Software Foundation. It includes a majority of semantic web technology standards, such as RDF API, SPARQL query language, and Ontology API. • Allegro graph: It was created to satisfy World Wide Web Consortium guidelines for the RDF, so it is appropriately viewed as a Relational Database Framework. It is a reference usage for the SPARQL rules. These rules are a standard query language for connected information, filling similar needs for RDF databases that SQL serves for relational database. Allegro graph is a closed source triple store which is intended to store RDF.

Out of these listed tools we are using Jena because apache

Jena supports all operating system like Linux, Windows • CSV2RDF: CSV2RDF is a java-based tool that is used for converting CSV files into generic triplets or RDF. For achieving this it uses CONSTRUCT query to create mappings.

Out of listed above we are using CSV2RDF because it is easy to use and we are taking product catalogues in CSV format in our proposed framework. It just takes CSV files as input and converts them into RDF format data. It is a java-based application and we also have implemented our framework in java hence we have used CSV2RDF tool

5. Implementation

In this section we are explaining steps followed while implementing each module of proposed framework to facilitate product searching. As we already discuss we have divided the task into two modules, knowledge modeling and knowledge consumption.

5.1. Knowledge Modelling

etc., with a Java Virtual Machine. It is an open source providing diferent API’s to extract data from and write to RDF graphs. The graphs represented as an abstract model. A model can be sourced with data from files, Databases, URLs or combination of these. accessible as an Apache 2 Licensed people group edition. The IDE provides mix with manufacture/bundling devices. IntelliJ has a lot of features and it is designed for making large projects. It supports languages like Kotlin, Java script, SQL etc.

If you want to fetch a correct and accurate information

4.4. Integrated Development according to your need than your knowledge base i.e., Environment where the information is stored should be managed properly. We are using the domain ontology for storing our Many IDE are used in computer programming. They information. The ontology is structured vocabulary that contain a base workspace and an extensible plug-in sys- defines the knowledge as concept[ 12 ]. Ontologies are tem for managing the environment. Few of them we are framework which is used as bases for linked data modlisting here:. eling because of its ability to explaining relationships and interconnectedness between them. More specific • Eclipse: It is broadly utilized among business organi- knowledge can be captured because of the semantic relazations to make amazing applications for programming tions of instances with their attributes[ 10 ]. This module improvement, yet for diferent enterprises, for example, further has two components: schema creation and data banking, automobiles, clinical and space investigation. integration.

Since it is coded in Java, Eclipse underpins most stages and working frameworks. Nonetheless, Eclipse isn’t • Schema Creation: First step is to determine the scope constrained to Java. It also allows us for using several and purpose of domain ontology. At this stage we languages by using their components of plugins. determine the scope and purpose of creating ontology. Before making ontology our purpose of mak• NetBeans: It is an integrated development environ- ing it should clear. Our ontology based on electronic ment (IDE) for Java. It allows us to create applications products provide us some semantic information about from a lot of small units of programs segments called the products that improves product searching results. modules. It supports various extensions of other pro- Therefore, the more semantic information about the gramming languages like PHP, C, C++, HTML5, and product should be provided in order to have more and JavaScript. This is an open-source integrated develop- more accurate information. Second step is collecting inment environment. formation about domain. When the idea about domain information and the domain knowledge is collected, • IntelliJ: It is an IDE coded in Java for creating computer programming. It is developed by JetBrains and is now we are able to create a good ontology which describes our requirement correctly. Third step is defining the classes and topology. There are three methods to design the classes. First one is top-down method- In this first we create general class, then we create specific classes as per use. Second one is bottom-up method In this first we create specific classes and then we create general classes. Third is the combination method In this we create most important classes first and then move towards general classes. Last step of schema creation is defining the properties- only class cannot give all the information therefore we have to define the properties of classes. We should define properties of general classes because all its properties are inherited by its sub-classes.

We created an ontology about computers and laptops where we have defined following classes, object properties, data properties and structure of ontology as shown in images below:

Initially we added seven instances i.e. details of five laptops/desktop manually using protege. Therefore, we have seven instances in our knowledge graph. We

have created an interface to add instances directly to our knowledge graph i.e. in our RDF file. Retailer will upload laptop/computer catalogues in CSV format, thereby storing product catalogues in our knowledgebase for consumption (query) by the end user. Once • Data Integration: At last we will add instances according to the schema we have created. For populating we first select a class then we create an instance for that and after we populate it as an instance to the category it belongs. When we populate instances on our knowledge model a knowledge graph is created. We can convert any format of data like csv, json, json-ld etc, according to our schema using RDFizer. we receive the catalogues, CSV2RDF a java-based application which use Jena API will convert it from CSV to RDF format by creating mapping between base URI of our ontology and CSV column. There are two scenarios, first is that we create separate RDF file for data provided by each retailer and after that we merge all the files into a single RDF file and second being that all the data can be annotated into a single RDF file as per the requirements of the application. We have annotated all the data to a single RDF file.

Every time retailer uploads their product catalogues it

is annotated semantically according to data model and all he instances added into knowledge graph. There are few new instances are present in updated knowledge graph shown in Figure 7.

5.2. Knowledge Consumption

In knowledge modeling module every information about

product catalogues are stored in machine understandable form in triple store. In knowledge consumption module we query the information present in triple store and fetch out the products that matches with user requirements.

RDF files generated after data integration module are stored in a triple store, Jena tdb. Jena tdb provides APIs to perform SPARQL Queries.

An interface is created for end users to enter their specifications. On clicking search button SPARQL endpoint will generate a SPARQL query shown in Figure 8. which execute on knowledge graph and products instances matches with these specifications are fetched.

For example user is looking for a laptop that has RAM = 4GB and has PROCESSOR = Intel Core i3. Two instances of laptops that are present in knowledgebase matches with these specifications i.e. Dell Vostro 15 3568 and Dell Inspiron 5568(Z564304SIN9) 2 In 1 laptop are recommended to end user.

6. Conclusion and Future Work

As we know the success of any e-commerce store de

pends on the database that how they have stored the information and how eficiently from it data can be retrieved. Products catalogues are in heterogeneous format. To organize information in a better way this work uses a knowledgebase which contains a semantic model that store information in a common format and in machine readable form. The challenge of eficiently using the information from the bulk of stored data has been made possible with the help of semantic web. The model helps us to get better result because semantic relationship helps the search engine to understand intent of user query in the same manner as human brain and hence products match with user requirements are fetched. In future we will evolve our ontology to add more classes about other electronic devices and publish our ontology on web and for input we are going to use web pages from which data is linked to our model and the instances will populate be populated into knowledge graph. [7] SELECTUSA retailpage, https://www.selectusa.

gov/retail-services-industry-united-states

References

[1] Ristoski , Petar, Petar Petrovski, Peter Mika , and Heiko Paulheim . "A machine learning approach for product matching and categorization." Semantic web 9 , no. 5 ( 2018 ): 707 - 728 .

[2] Rif 'at, Muhammad, Rahmad Mahendra, Indra Budi, and Haryo Akbarianto Wibowo. "Towards product attributes extraction in Indonesian e-commerce platform . " Computación y Sistemas 22 , no. 4 ( 2018 ).

[3] Shah , Kashif, Selcuk Kopru , and Jean David Ruvini. "Neural network based extreme classification and similarity models for product matching." Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , Volume 3 ( Industry Papers). 2018 .

[4] Al-Taie , Mohammed Zuhair, Siti Mariyam Shamsuddin, and Joel Pinho Lucas. "Predicting the Relevance of Search Results for E-Commerce Systems." Int. J. Advance Soft Compu. Appl 7 .3 ( 2015 ).

[5] Zhang, Liyi, Mingzhu Zhu , and Wei Huang . "A framework for an ontology-based e-commerce product information retrieval system . " Journal of Computers 4.6 ( 2009 ): 436 - 443 .

[6] Guha , R. , R. McCool , and E. Miller . "Semantic search . WWW'03: Proceedings of the 12th international conference on World Wide Web." ( 2003 ): 700 - 709 .

[8] Salatino , Angelo, Thiviyan Thanapalasingam, Andrea Mannocci, Francesco Osborne, and Enrico Motta . "Classifying research papers with the computer science ontology . " ( 2018 ).

[9] Peis , Eduardo, JM Morales del Castillo, and Juan A . Delgado-López . "Semantic recommender systems. analysis of the state of the topic . " Hipertext. net 6 . 2008 ( 2008 ): 1 - 5 .

[10] Kutiyanawala , Aliasgar, and Prateek Verma . "Towards a simplified ontology for better e-commerce search." arXiv preprint arXiv: 1807 . 02039 ( 2018 ).

[11] Jain , Sarika. "Exploiting Knowledge Graphs for Facilitating Product/Service Discovery." arXiv preprint arXiv: 2010 . 05213 ( 2020 ).

[12] Ontotext , https://www.Ontotext.com/knowledgehub/ fundamentals/what-are-ontologies/

[13] BrainSINS, https://www.brainsins.com/en/blog/ ai -powered-semantic-search-increase-ecommerce/ conversion/4526

[14] Marketing charts ,https://www.marketingcharts. com/digital/seo-83805

[15] Statista https://www.statista.com/topics/871/online/ shopping/

[16] Aslam , Zulqurnan. "Semantic Web Mining in E-Commerce Websites . " International Journal of Computer Applications 137.2 ( 2016 ).

[17] Minch , Arba. "Engineering Semantic Web for E-Commerce Business Intelligence: a Bilingual EEPS Ontology Model."PhD diss ., Arba Minch University, 2014 .

[18] Malve , Ankita, and P. P. M. Chawan . "A comparative study of keyword and semantic based search engine." International Journal of Innovative Research in Science, Engineering and Technology 4 .11 ( 2015 ): 11156 - 11161 .

[19] Benjamins , Victor Richard, Jesús Contreras, Oscar Corcho, and Asuncion Gomez-Perez. "The six challenges of the Semantic Web . " ( 2002 )..

[20] Sonika Malik , Sarika Jain, “ Semantic OntologyBased Approach to Enhance Text Classification” , CEUR Workshop Proceedings , vol 2786 , pp. 85 - 98 . International Semantic Intelligence Conference, Delhi, India. 25 - 27 Feb 2021 .

[21] Kutiyanawala , Aliasgar, and Prateek Verma . "Towards a simplified ontology for better e-commerce search." arXiv preprint arXiv: 1807 . 02039 ( 2018 ).

[22] Necula , Sabina-Cristiana , Vasile-Daniel Păvăloaia, Cătălin Strîmbei, and Octavian Dospinescu . "Enhancement of e-commerce websites with semantic web technologies . " Sustainability 10 , no. 6 ( 2018 ): 1955 .

[23] Rudman , Riaan, and Rikus Bruwer . "Defining Web 3.0: opportunities and challenges." The Electronic Library ( 2016 ).

[24] Jain , Sarika. Understanding Semantics-Based Decision Support . CRC Press, 2020 .