botego chatbots general information
TRANSCRIPT
Adding questions and answers and delivering them
• How are questions and answers added to the system?
• Questions and answers are usually extracted from sources such as:
• Call center transcripts
• Support e-mails
• Support tickets
and they’re either manually entered into a form, or automatically imported from a CSV file.
• Either way, the algorithm processes these questions and answers, removes the stop words, finds the roots of the
remaining words based on our proprietary technology, and associates them as keywords, with the relevant
question.
• How does the system recognize different variations of the question?
• Synonyms are entered as keywords in order for the system to recognize variations. i.e. Keywords such
as car, automobile and vehicle can make a question discoverable through each of these words.
• How does the algorithm work?
• Our algorithm is similar to tf-idf (stands for term ‘frequency, inverse document frequency’) which is
widely used for information retrieval. User input is taken through a process where stop-words are
removed and roots of the remaining words are found. They are compared with the keywords in the
database, and answer associated with the most relevant keyword is delivered to the user, based on a
pattern-matching score algorithm.
• Which languages are supported?
• Currently, following languages are supported through various NLP libraries that can perform stemming
in those languages:
• English
• French
• Spanish
• Portuguese
• Italian
• Romanian
• German
• Swedish
• Norwegian
• Danish
• Russian
• Finnish
• Turkish
• Can it retrieve information over a web service?
• An answer can contain a variable which is supposed to be replaced just before the answer is delivered
to the user. Actual value of the variable can be retrieved from a web service. A web service integration
task is involved with such a feature.
• i.e. If the price of a specific stock is asked by the user, this information is pulled from a source, added to
the answer and delivered to the user.
• What are the available formats for an answer? Can it be text or a form?
• Questions and answers are in the HTML form. Bold or italic text, a hyperlink, a table, an image, or even
a video can be included in the answer, making them an ideal form to deliver guided visual information
elements.
• How are the questions and answers organized?
• Questions and answers are listed on the dashboard. They may be grouped based on context, for easy
content management.
• How many questions and answers are required? Is there a technical limitation?
• We can’t talk about a ‘one size fits all’ approach in terms of number of questions. Typically, 80% of the
user questions represent only 20% of the knowledge base. So, we recommend to cover that 80% first,
for an efficient operation.
• Regarding limitations, a typical customer with 10,000 question-answer pairs in the database serving
more than 13,000 employees requires a virtual server with 8GB-RAM and a 3GHZ processor to deliver
more than 50,000 questions per day, in under one second.
• Can we develop a conversation scenario for a guided conversation?
• Various questions can be ‘linked’ to each other to form a conversation scenario. i.e: A scenario about a
broken phone can start with asking the user what model the phone is. And based on that answer,
various other questions follow.
• Linking two questions to each other is pretty easy using the ‘link’ feature on the dashboard. Simply point
the question you want to link to, and user will be presented with a clickable link to answer that question.
Context-sensitivity
• Is the system context-sensitive? Can it remember previous answers?
• Context-sensitivity is established manually. Thanks to question linking feature, the answer given to a
certain question can vary based on user’s answer to the previous question.
• Can the system recognize users? Can it deliver custom answers based on their profile?
• A parameter associated with a profile can help us link the user to a specific question and answer group.
So, answers delivered to that user will be only from that answer pool. A premium user will only get
answers from the premium user content.
• What is a session?
• A session technically never ends, and a conversation log is limitless. A valid session information will
continue the previous session. Otherwise, a new session will be started.
The technology
• What are the technologies that you have used? Do they require purchase of any licenses?
• We use open source technologies with no licensing cost. Python, PHP, mySQL, mongoDB, redis,
rabbitmq, celery and other complementary systems have been used.
• What is the reason of your choice?
• Since the application is web based, programming technologies such as php, mysql which are widely used
for web applications are chosen. We also have considerable know-how about these technologies. As a
result, choosing them help us develop more quickly and with less cost.
• Do they run on Windows?
• Application server doesn’t run on Windows, but there’s no limitation for the end user chat interface.
• Which platforms are supported?
• Chat interface connects to the chat server over RESTful web services. So,chat interface can be
implemented on any technology or platform that has access to HTTP web services. Our default chat
interface is web-based and works on popular websites such as Firefox, Chrome, Safari, Internet
Explorer 7+.
• On which platform does the content management interface work?
• Content management interface is web based and works on Firefox, Chrome, Safari and Internet
Explorer 10+. Currently, we don’t provide mobile access to the content management interface.
• Do you offer API’s?
• Yes, we do offer API’s written in PHP and Python. Functions such as initiating a chat, sending a
message, reading a message can be done through the API.
Installation & Hosting
• How is the application hosted?
• Some of our customers prefer their applications to be hosted on Botego’s Amazon cloud servers,
whereas banks and telecom companies usually prefer their on-premise servers due to mission critical
functions. In this case, we install the server and the application either via remote connection, or on-
premise.
• What are the server requirements?
• Depending on the conversation load on the application and availability-performance requirements of the
customer, number of servers vary from one to eight.
• A server with 8 MB’s of RAM and a 3GHz processor will be able to handle 4000 questions per hour.
New servers can be added for backups and to form a failover structure.
• Is it scalable?
• Since queries used for the pattern-matching feature takes place on the database, heaviest load is
usually on the database server. So, database layer can be scaled by using sharding and clustering
approaches.
Transfer to the live agent
• User can be transferred to a live agent queue, based on a pre-determined condition such as failing to find a
relevant answer for the question three times in a row. To do so, the application displays a link for the user to click,
or a message such as ‘Please wait as I forward you to a live operator’.
Reports
• Daily, weekly or monthly reports can be automatically generated for data such as number of questions answered,number of conversations, most frequently asked questions, and many others.
Botego Inc.115 E 23rd St New York 10010 NY
646 932 96 04