Superagent: A Customer Service Chatbot For E-Commerce Websites

Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 4

SuperAgent: A Customer Service Chatbot for E-commerce Websites

Abdullah
119A1042
C2

Abstract: -
Conventional customer service chatbots are usually based on human dialogue, yet significant
issues in terms of data scale and privacy. In this paper, we present SuperAgent, a customer
service chatbot that leverages large-scale and publicly available ecommerce data. Distinct from
existing counterparts, SuperAgent takes advantage of data from in-page product descriptions as
well as user-generated content from ecommerce websites, which is more practical and cost-
effective when answering repetitive questions.

Introduction: -
Customer service plays an important role in an organization’s ability to generate income and
revenue. It is often the most resource-intensive department within a company, consuming
billions of dollars a year to change the entire perception customers hold. Support staff spend a
lot of time answering questions via telephone or messaging applications to make sure are
customers satisfied with their business. This traditional customer service has two problems:
First, staff usually receive repetitive questions asked by a variety of customers, which can be
cost-effectively answered by machines. Second, it is difficult to support 7×24 services, especially
for most nonglobal businesses. Therefore, chatbots can be a great way to supplement customer
service offerings since they are more economical and indefatigable, and free up support staff to
answer much higher value queries.
Literature Survey: -
SrNo Paper Name Author Reference
.
1. Neural Machine Dzmitry
Translation by Bahdanau, Neural machine translation is a recently proposed
approach to machine translation. Unlike the
Jointly Learning to Kyunghyun traditional statistical machine translation, the
Align and Translate Cho, Yoshua neural machine translation aims at building a
Bengio single neural network that can be jointly tuned to
maximize the translation performance. The models
proposed recently for neural machine translation
often belong to a family of encoder-decoders and
consists of an encoder that encodes a source
sentence into a fixed-length vector from which a
decoder generates a translation.
2. Learning deep Po-Sen Latent semantic models, such as LSA, intend
structured semantic Huang, to map a query to its relevant documents at
models for web Xiaodong He, the semantic level where keyword-based
search using Jianfeng Gao, matching often fails. In this study the authors
clickthrough data Li Deng, strive to develop a series of new latent
Alex Acero, semantic models with a deep structure that
Larry Heck. project queries and documents into a
common low-dimensional space where the
relevance of a document given a query is
readily computed as the distance between
them.
3. Quantile Regression Nicolai Random forests were introduced as a
Forests Meinshausen machine learning tool in Breiman (2001) and
have since proven to be very popular and
powerful for high-dimensional regression and
classification. For regression, random forests
give an accurate approximation of the
conditional mean of a response variable. It is
shown here that random forests provide
information about the full conditional
distribution of the response variable, not only
about the conditional mean.
4. GloVe: Global Vectors Jeffrey The authors analyze and make explicit the model
for Word Pennington, properties needed for regularities to emerge in
Representation Richard Socher, word vectors. The result is a new global
Christopher D. logbilinear regression model that combines the
Manning advantages of the two major model families in the
literature: global matrix factorization and local
context window methods. Their model efficiently
leverages statistical information by training only
on the nonzero elements in a word-word
cooccurrence matrix.

Module Diagram: -

As the figure shows, when the product page is first visited, SuperAgent crawls the html
information and scrapes data from the webpage.
After that, given an input query from a customer, different engines are processed in parallel. If
one of the answers from the first three engines has high confidence, the chatbot returns with
the answer as the response. Otherwise, the chit-chat engine will generate a reply from the
predefined permitted response sets.
The chit-chat engine is mainly designed to reply to greeting queries such as “hello“ and “thank
you”, as well as queries that cannot be answered by the previous three engines, such as “you
are so cute”. However, general chit-chat engines tend to be topic-deviated so that the replies
may be irrelevant. To avoid such deviations, we follow the smart reply approach for email reply
suggestions to predefine a permitted response set. Formally, the chit-chat model is an
attention-based LSTM seq2seq model trained on twitter conversation data (˜43 million query-
reply pairs).

Results: -
The bot selects the 5 million most frequent non-duplicate short replies (less than 15 words) as
the permitted response set, most of which are greetings and common replies. The end-to-end
perplexity of the seq2seq model is 16.9, which is similar to Kannan et al.’s result. The engine’s
output is very topic-coherent, which is shown as follows:
Q: hello
R: hey how are you?
Q: thank you
R: you’re very welcome sir
Q: you are so cute
R: u r more

Conclusion: -

Compared to conventional customer service chatbots, SuperAgent takes advantage of large-


scale, publicly available, and crowd-sourced customer data. In addition, SuperAgent leverages
state-of-the-art NLP and machine learning techniques, including fact QA, FAQ search, opinion-
oriented text QA, as well as chit-chat conversation modeling. Usability analysis shows that
SuperAgent has improved the end-to-end user experience in terms of online shopping. It is
more convenient for customer’s information acquisition especially when a product page
contains too much user-generated content.

You might also like