= !OfficeBot: joint work of NLPC MU and Konica Minolta =

student R. Sabol

a Slack bot that supports teamwork

Project web: https://nlp.fi.muni.cz/projects/officebot/ [[BR]]
Project repository: [source: /nlp/projekty/officebot/trac-git]

tasks:
* detect channel activity
  * active/non-active (displayed as bold/normal typeface)
  * type of activity (conversation = several mutually related posts, individual posts)
  * language of the activity (!Czech/English, short messages/long posts ...)
  * sentiment of the activity (possible flamewar, emoji-based classification?)
* summarize activity for a particular channel
  * from a particular moment track all messages
  * cluster conversations (based on time span and/or topic)
  * detect keywords
  * highlight most important messages or part of messages
* summarize activity for all channels
* suggest joining a conversation based on user preferences (favorite topics, languages, sentiment ...)
* suggest moving part of a conversation into external applications
  * transfer to calendar, trello, private channel (talk to myself)


data:
* publicly available dataset https://github.com/houstondatavis/slack-export
* KM data (to be provided after filtering out sensitive information + solving legal issues)

language:
* English (needed to filter out non-English texts)
* probably needed tools for "internet language" (e.g. abbreviation expansion, emojis)

== Plan ==

|| Time period  || Student Activity || KM Activity ||
|| M1 || \
{{{#!td
study the data

study Slack functionalities

[[ChatbotFunctions|bring ideas about what can be done]]
}}}
{{{#!td
resolve legal issues concerning KM conversations data
}}}
|----------------
|| M2 || \
{{{#!td
create the basic chatbot:
* listen on different channels
* detect channel activity
* quantify channel activity
}}}
{{{#!td
run the chatbot on KMLE Slack

start collecting feedback
}}}
|----------------
|| M3 || \
{{{#!td
study NLP techniques/implementations that can support smart features for the chatbot

extend the chatbot with language detection

propose a evaluation for the new feature

cluster conversations based on time span

}}}
{{{#!td
}}}
|----------------
|| M4-M5 || \
{{{#!td
extend the chatbot with keyword detection using TF-IDF or another "simple" technique

study the specifics of the Slack conversations

cluster conversations based on keywords

}}}
{{{#!td
}}}
|----------------
|| M5-M6 || \
{{{#!td
study techniques for text summarization

extend the chatbot with summarization

}}}
{{{#!td
provide feedback for keyword detection
}}}
|----------------
|| M7 || \
{{{#!td
extend the chatbot with sentiment analysis based on words/emojis

suggest joining a conversation based on
keywords/topics/participants/sentiment...
}}}
{{{#!td
provide feedback for text summarization
}}}
|----------------
|| M4-M8 || \
{{{#!td
extend the chatbot with connectors to applications such as Trello, calendar etc. 	
}}}
{{{#!td
provide feedback for sentiment analysis
}}}
|----------------
|| M9 || \
{{{#!td
evaluate and deliver the application 	
}}}
{{{#!td
}}}

**Documentation**

Scopes for .history API call:
* channels.history: collects history only from SLACK public channels
* groups.history: same as with channels, but this one works only for private channels
* im.history: collects chat history from private conversations
* mpim.history: collects chat history from multi-party conversations
* conversations.history: universal scope for all 4 API calls mentioned above, it´s functionality depends on which of those 4 scopes is chatbot allowed to use. For example: if chatbot is allowed to view groups and channels but not allowed to view private messages and multi-party conversations, those API-calls get rejected automatically without limiting other scopes

List of potentionally useful information that can be obtained with SLACK API calls:
* channels.history:
  * list of all messages, events
  * timestamps of these messages
* users.info:
  * name, real name, team name, e-mail
  * info about user being admin, owner, bot, restriced/ultra-restricted
* channels.info:
  * all members
  * scope of the channel (private, public, mpim..)
  * latest message
  * number of unread messages
  * topic
  * previous names


Potential ideas:
Hard statistics:
* using timestamps to compute ratio of messages in certain time interval
* using join time/leave time to compute user´s time online over certain period of time
* limit channel history checking to certain time interval
* limit channel history to certain scope of users (people from same team, admins...)
* finding the moment where conversation had most messages for certain time interval
Soft statistics:
* identify dialogs
* statistics for dialogs (length, participants..)
* analysis of dialogue - keywords, named entities, topics