International Journal of Computer Engineering and Applications, Volume XI, Special Issue,
May 17, www.ijcea.com ISSN 2321-3469
Aditya Deshpande, Alisha Shahane , Darshana Gadre, Mrunmayi Deshpande, Prof. Dr. Prachi M.
desks, e-commerce and so on. In the proposed system the user input is given to the semantic
mapper, which maps the input to semantic elements. These elements are given to conflict mediator
in order to resolve conflicts by having further conversation with user, and are passed to the topic
navigator. If there are no conflicts then the elements are directly given to topic navigator which
finds the appropriate answer in the information repository.
This answer is given to the response
generator for generating a natural language response to be given to the user. Three kinds of chatbot
namely Basebot(contains converstional knowledge), Domainbot( contains domain related faqs),
Repbot(hybrid) were used to tedt the efficiency of the proposed system. It was found that a hybrid
chatbot yielded best response satisfaction rate and least topic switching rate.
According to the
results, conversational knowledge Basebot combined with topic specific knowledge should be
adopted for future applications .
Another approach of implementing a chatbot is the smart answering OCR based chatbot.
This approach uses the Optical Character Recognition technology(OCR), Overgenerating
transformations and ranking algorithm and Artificial Intelligence Markup Language(AIML). OCR
technology is a mechanism of converting a scanned document, images of hand written text into
machine encoded text. Overgenerating transformations and ranking algorithm generates logically
equivalent questions from source sentences. AIML is an XML dialect for creating natural language
software agents. The proposed system has three phases Plain text extraction, Question Generation
and Question and Answers. Plain text is extracted from pdf documents or images using OCR
technology. Questions are generated from the extracted text via the overgenerating transformations
and ranking algorithms. The question-answer pairs that are generated are stored as the chatbot
knowledge using AIML. A pattern matching algorithm is used to match the user input to the data
stored in AIML. The corresponding responses are given to the user. This approach provides an
efficient way of converting documents into the chatbot knowledge. This system can be used in call
center services and educational field for answering frequently asked questions .
An application of chatbots lies in the field of E-business and e-commerce. The main
problem that almost every e-business model currently faces is that of quality customer service in
the least amount of time. As a solution to this problem, a solution is proposed by Thomas N T that
consists of a chatbot system to generate immediate responses, which is a combination of AIML and
LSA . Template based questions and greetings are answered by using AIML and other general
questions are answered by using LSA. The user query is first passed to the AIML block, which
checks if the query is template based. If yes, then a pattern based answer is generates as response.
Otherwise, the query is routed to the LSA block where trained data is required to match the user
query with expected output. The FAQs in any particular e-business domain is used for training the
model. The FAQ is made using online data from the internet. The FAQ corpus passes through a
series of steps beginning with tokenization where tokens are formed. Then stop word removal is
performed by using Porter stemmer algorithm. After this, a word-document matrix is generated and
then SVD is computed. Cosine similarity is used to evaluate result with minimum distance from
user query and this result is generated as the response. User queries are stored in HBase and AIML
database is updated to improve answers to template based questions. The model achieved 0.97
precision and LSA based questions gave correct responses .
When the user provides insufficient information to answer his query successfully, the
chatbot needs to be inquisitive, that is it must proactively ask the user questions in order to mimic a
more natural human interaction. This approach details the implementation of such an inquisitive
chatbot which recognizes missing data from a query and probes the user to obtain the same in order
to answer his query. In the existing chatbots, the chat engine uses pattern-matching algorithms to
search the knowledge base for. ALICE engine uses AIML as a knowledge base to stores a set of
predefined queries and its variants. In order to make these hard-coded answers dynamic, we