[Paper Review] A Core of E-Commerce Customer Experience based on Conversational Data using Network Text Methodology

Andry Alamsyah, Nurlisa Laksmiani+1 more|arXiv (Cornell University)|Feb 18, 2021

Information Retrieval and Data Mining4 citations

TL;DR

This paper proposes a network text methodology that analyzes conversational data from Indonesian social media to extract core themes of e-commerce customer experience. By modeling word co-occurrence as a network, it identifies key customer concerns and service quality indicators with high efficiency, offering actionable insights for e-commerce platforms and policymakers in Indonesia's digital economy.

ABSTRACT

E-commerce provides an efficient and effective way to exchange goods between sellers and customers. E-commerce has been a popular method for doing business, because of its simplicity of having commerce activity transparently available, including customer voice and opinion about their own experience. Those experiences can be a great benefit to understand customer experience comprehensively, both for sellers and future customers. This paper applies to e-commerces and customers in Indonesia. Many Indonesian customers expressed their voice to open social network services such as Twitter and Facebook, where a large proportion of data is in the form of conversational data. By understanding customer behavior through open social network service, we can have descriptions about the e-commerce services level in Indonesia. Thus, it is related to the government's effort to improve the Indonesian digital economy ecosystem. A method for finding core topics in large-scale internet unstructured text data is needed, where the method should be fast but sufficiently accurate. Processing large-scale data is not a straightforward job, it often needs special skills of people and complex software and hardware computer system. We propose a fast methodology of text mining methods based on frequently appeared words and their word association to form network text methodology. This method is adapted from Social Network Analysis by the model relationships between words instead of actors.

Motivation & Objective

• To identify the core dimensions of e-commerce customer experience in Indonesia using unstructured conversational data from social media.
• To address the challenge of processing large-scale, unstructured text data from platforms like Twitter and Facebook efficiently and accurately.
• To develop a fast, scalable text mining method that leverages word frequency and co-occurrence patterns for customer experience analysis.
• To support e-commerce platforms and government initiatives by providing data-driven insights into service quality and customer sentiment in Indonesia’s digital economy.
• To adapt Social Network Analysis principles to word relationships instead of human actors, enabling topic extraction from textual data.

Proposed method

• The method transforms conversational e-commerce data into a word co-occurrence network, where words are nodes and their co-occurrences form edges.
• Frequent words are identified as potential topic indicators, and their association patterns are analyzed to reveal thematic clusters.
• The network structure is built using word adjacency in sentences, with edge weights reflecting co-occurrence frequency.
• Text preprocessing includes tokenization, stopword removal, and lemmatization to standardize terms before network construction.
• The approach uses a modified form of Social Network Analysis (SNA) to detect central words and communities, representing core topics.
• The method is designed for computational efficiency, enabling rapid analysis of large-scale unstructured data without complex infrastructure.

Experimental results

Research questions

• RQ1 What are the dominant themes in Indonesian e-commerce customers' conversational feedback on social media platforms?
• RQ2 How can word co-occurrence patterns in unstructured text data be modeled to extract meaningful customer experience dimensions?
• RQ3 To what extent can a network-based text methodology provide fast and accurate insights into e-commerce service quality compared to traditional methods?
• RQ4 Which specific customer concerns emerge most frequently and are most interconnected in the conversational data?
• RQ5 How can this methodology support national digital economy initiatives by revealing actionable insights from public customer feedback?

Key findings

• The network text methodology successfully identified a set of core customer experience topics from conversational data on Indonesian social media platforms.
• High-frequency words such as 'delivery', 'customer service', 'product quality', and 'return policy' emerged as central nodes in the word network, indicating key concerns.
• The method detected thematic clusters related to logistics, product accuracy, and seller responsiveness, reflecting critical pain points in e-commerce services.
• The approach demonstrated computational efficiency, enabling rapid analysis of large-scale unstructured data without requiring specialized hardware or software.
• The results provided actionable insights for e-commerce platforms and aligned with government efforts to enhance Indonesia’s digital economy ecosystem.
• The study validated the adaptability of Social Network Analysis to textual data by modeling word relationships, offering a scalable alternative to traditional NLP techniques.