[Pycon] [new paper] "Kathleen Siminyu" - Real-Time Auto-Tagging of Chat Dialogues for Efficient Client Relations and Support Operations

[info a pycon.it]

Title: Real-Time Auto-Tagging of Chat Dialogues for Efficient Client Relations and Support Operations
Duration: 240 (includes Q&A)
Q&A Session: 0
Language: en
Type: Training

Abstract: Instant Messaging(IM) is a type of online chat that offers real-time text transmission over the Internet. A use case for IM that is increasingly gaining popularity is that of businesses using it to offer services as well as support to their clients.

Identifying as early as possible in an IM conversation where/who to route communication to, for a business, would assist in streamlining client relations and support operations for maximum efficiency. The ability to do this becomes more crucial as the customer base of a business grows and the distribution of teams that handle different products widens.

This use case introduces the problem of handling unstructured dialogue data. The messages involve context switching. Beginning as a dialogue between two people, once the client’s request or query has been determined a third party more qualified to handle the situation may be added to the conversation either explicitly, their presence is known to the client, or implicitly. The conversation may also be reassigned to a completely different agent or perhaps a suggestion made by a bot solves the client’s query without the need for human intervention. 

The dataset comprises 20,000 historical conversations between the clients of Africa’s Talking, a communication service provider in several markets in Africa, and the client relations and support agents. We infer tags for these conversations by using a feature extraction technique known as the Term frequency-inverse document frequency (TF-IDF). TF-IDF is a feature vectorization method widely used in text mining to reflect the importance of a term to a document in the corpus. The most important terms of each conversation are regarded as tags. 

The resultant tagged dataset was then used to train a predictive classification model that can auto-tag real time conversations, determining the relevant issue or product, thus routing them to the most qualified agent.


Tags: [u'nlp', u'Artificial Intelligence']