- - 6 min read
Cobot Architecture and Applications Overview
- - posted on March 19, 2019
Introduction to Cobot
This is the age of Artificial Intelligence and Chatbot. Mostly chatbot which is used or previous chatbot which was used considered to be rule-based generally, but the quest of making them intelligent has begun. The Integration of Chatbot with cognitive technologies is a step toward this direction. Consider a scenario that in a company an employee requested to HR to make a file of the employees who come late at work with the timing at which they came. This is not a typical task when the count of the company is somewhere let’s say 50 employees but what if the number is near about 500 employees and what if this request can be passed to a bot through text/voice and it can fulfill the request irrespective of the size and number of employees.
This can be done by giving the cognitive power to bot. The bot also requires a conversational framework. This conversational framework should address the following key points -
- Question Analysis - Includes Natural Language Understanding (NLU), Resolution of the context
- Personalization - Includes knowing the user, knowing the sentiments of the user and responding according to it and making recommendations
- Knowledge Representation - Knowing the domain of the problem to be addressed and acknowledging using the backend model, should have reasoning power too.
- Dialog Authoring - Generation of the dialogs in a semi-automated manner using various support artifacts.
So before diving deep into the concepts of COBOT. Let’s understand some terminologies related to it. The main requirements of COBOT are cognitive analytics and Chatbot.
What are Chatbots?
A chatbot is a program that recreates a genuine discussion that you would usually have with a customer service representative. The utilization of bots has many possibilities, from practical to fun, and it can be used in any real-time messaging platform, such as Facebook Messenger, Telegram, Slack, instant messages, and so forth. There are three types of chatbots -
Bot as Introducer -
- For Email Marketing
- For Social Media
- For Video
- For Search
Bot as Influencer -For a bot to be a successful Influencer you must always consider how they can be helpful and how they can add value.
Bot as closer -Leverage bots to remove any friction and make the experience super relevant and Empathetic.
What is Cognitive technology?
Cognitive Automation is an intelligent and smart automation method based on Artificial Intelligence used to replicate human capabilities including simpler mental processes like thinking, reasoning, decision making, and other activities, providing a bridge for linking human consciousness and the static logic of computing.
The collaboration of these two technologies gives birth to COBOTs.
Master Architecture - Description
- The architecture contains three main parts Introducer bot, Influencer bot, and Closer Bot.
- Introducer bot will act like a prime interface which will take a query and has a task to respond accordingly.
- Influencer bot will serve as the front-end to Introducer bot or to a user for making and handling their queries.
- Closer bots will act as domain experts. So each domain will have a bot representation. These bots can handle their domain-specific query using a defined ontology graph, and they will provide a relevant solution.
- The whole framework can be triggered when a client put the request of an error encounter, and he/she summons the introducer bot.
- In the case where a query comes which is beyond the scope of these bots combined, the query will be passed to the human expert of that specific domain.
Cobot's Working Architecture
Registration of the user - If the user is not registered with framework than the system will ask to register and if the user is already logged than the structure will move toward the Identification of the Domain and understanding the problem accordingly.
Identification of the Domain - The influencer bot start the procedure of the identification of the Domain using user query or from the data of the past encounter saved in the database, enabling the personal touch.
Problem and its related Entity Extraction - Next step is to identify the problem from which the Key and Intent belong to the user query identified.
Routing the Query generated by the problem - From here the query and its attributes are passed to the domain-specific bot or the Information Retrieval (IR) engine. If the query belongs to which require general troubleshooting only then it will be handled by Influencer Bot using its IR engine and will provide the relevant solution URLs. If the query belongs to a specific domain then, it will be passed to the corresponding closer bot which is expert in that particular domain.
Extraction of the solution to the query for providing a solution a domain expert also use the query received by the Influencer bot and use it predefined ontology in which it checks if more equation is required on the query and asks the question using child node and node match procedure the user requests this question and on the basis of it solution or response is passed to the Influencer bot and then to the user through user bot.
Handling the Query which is beyond the scope of the framework - If a query is beyond the scope of Influencer bot as well as of the closer bot, then the query will be passed to the human expert of the related domain of the query.
Passing the response - Influencer bot collects the response from its IR or Closer bot or the Human agent and forwards to the user.
Architecture of the Key Bots
Key components -
- Conversational Interface
- Speech to text module
- Processing Model of the Bot
- Text to speech module
Key elements -
- Question Analysing Unit
- Information Retrieval Engine
- Personalization Module
Key components -
- Knowledge Representation using Ontology
- Dialog Authoring module
Description of the architecture
The main task of this bot is to interact with the user, taking the problem and providing the solution. In simple words, it acts as the front-end to both the remaining bots.
Although it also has its front end (conversational UI) and it is also divided into four different units. These units are Conversational Interface, Speech to text module, Processing Model of the Bot, Text to speech module.
It can take the problem from the user and problem-related data and pass to the influencer bot.
Influencer bot then takes further actions (whether to solve its influencer bot level or pass it to the closer bot (domain expert).
It then receives the response on the issue from Influencer bot (whether it a failure or success) and passes it to the user.
It should have the capability to understand the Natural Language query so that it can understand the question from the user and can pass the response either from the user or from the bot itself.
Firstly it divides the query into the user’s intent, key entities, and domain of the question it is identified using Machine learning trained model. Once the question is divided it tries to solve the query using its IR engine or if it fails it passes the query to its domain expert closer bot.
If the closer bot requires further information to solving the query, it passes that requirement to the user bot and from there to the user and handles the response.
Question Analysis -It is the process of analyzing the query in detail to determine what is the query’s intent, what are keywords and words from the user query, and what is the domain of the query.
Personalization - Personalization is the prime concern to be obtained while implementing a chatbot. Influencer bot achieves it by saving the query and its domain with the data of the user so that it can be used to give suggestions further. It is possible that an area has multiple systems or a user has asked query related to multiple domains, in such it shows the suggestions containing multiple query and multiple systems.
Closer bots are considered to be the domain expert bots. Each registered domain must have a bot under closer bots.
The functionality of closer bot has two main building blocks -Knowledge Representation and Dialog Authoring.
Knowledge Representation - Each closer bot depends on the knowledge graph known as ontology built from its domain knowledge source and historical ticket data which is used to establish the relationship between the query and its solution using domain and entities extracted. The figure below showing the example of ontolog.
Dialog Authoring - To solve a query closer bor need multiple interactions with the user sometimes. For that Dialog, Authoring is used which work by the ontology and handle the conversations.
- It can provide Technical Support at an enterprise level.
- It can handle the Initial level filtering in the Recruitment System of a company.
- It can be used to provide better-personalized experience to the customers/ clients online.
- It can also be used to handle online shopping customers (with recommendation engine and sentiment analysis) to provide better shopping experience and recommendations.