Segregation of chat sessions based on user query
Embodiments of the invention relate to chat and, more particularly, to determining an that is to be action taken based on the type of chat session. The resolution of the chat is categorized to decide the necessary steps taken and also to monitor the agent's performance. A chat filter extracts relevant portions of a chat session. The relevant factors are taken into consideration and scored based on the feature vectors. A model is built and the type of resolution is determined. An analysis of the chat session is then performed taking into consideration several factors.
1. A computer implemented chat analysis method, comprising:
providing a processor, said processor creating and using a model for classifying
resolutions in a chat session in phases comprising a model training phase, a model testing phase, and a model application phase;
during said model training phase, a first chat filter extracting relevant portions of a chat session;
said processor performing feature extraction on said relevant portions of said chat session to obtain feature vectors;
said processor determining said feature vectors according to category relevancy;
said processor scoring said feature vectors based on a probability or likelihood of classifying said feature into a particular one of a plurality of categories;
said processor ranking said categories by their scores; and
based on a most likely category, as determined by said ranking, said processor making a final assignment of a predicted classification of said feature;
during said model testing phase, a second chat filter extracting relevant portions of a chat session;
said processor analyzing said chat session;
said processor validating said model against labeled data; and
during said model application phase, said processor receiving user chat information comprising a user interaction during a chat session across a computer network from a computer chat system or a system for real time communication across said computer network;
said processor extracting unlabeled data from said user chat information;
said processor scoring said unlabeled data with said model;
during said chat session, said processor analyzing said user interaction with said model to predict a type of chat session;
said processor determining whether said chat session comprises an information based chat query or an action request-based chat query;
based upon said type determination, said processor determining an action to take in connection with said chat session;
when said chat session comprises an information based chat query, providing responsive information to said user; and
when said chat session comprises an action request-based chat query, performing a responsive action on behalf of the user.
2. The method of claim 1 , further comprising: said processor monitoring an agent's performance.
3. The method of claim 1 , wherein a chat session comprises any interaction between an agent and a customer via any of voice, textual chats, social networks, and forums.
4. The method of claim 1 , further comprising:
said processor providing a classifier that obtains data matrices from said chat sessions; and
said processor applying a classification algorithm to label said matrices.
5. The method of claim 4 , wherein said classifier provides a different number of maximum labels for each of a plurality of chat-related documents.
6. The method of claim 1 , further comprising:
said processor processing untagged data from one or more chat sessions through a relevant line extractor to remove any of extraneous and irrelevant lines, phrases, and tags from a chat transcript.
7. The method of claim 6 , further comprising:
said processor processing an output of said relevant line extractor, along with a model, to generate tagged data.
8. The method of claim 7 , further comprising:
said processor inputting said tagged data into a relevant line extractor.
9. The method of claim 8 , said relevant line extractor further comprising:
a feature extractor using language models to parse chat sessions and to extract any of relevant and important features from said chat sessions.
10. The method of claim 9 , further comprising:
said feature extractor using a natural language model to extract part of speech (POS) tags that are used as features.
11. The method of claim 9 , further comprising:
said processor providing an output of said feature extractor to a model building module to build machine learning models based on said tagged data.
12. The method of claim 11 , further comprising:
during said model execution stage, said processor using a model and untagged data to classify a chat session into any of an action-based chat session and an information-based chat session.
13. The method of claim 1 , wherein a chat session comprises text chat that is any of transcribed text and text that is obtained through any of social media and forums.
14. The method of claim 1 , further comprising: said processor classifying resolutions in a chat session.
15. The method of claim 1 , further comprising:
said processor identifying a mode of chat to determine when chat content is any of a text chat, a transcribed text chat, social media, and a forum.
16. The method of claim 1 , wherein said feature vectors comprise factors including any of weightages assigned to word count and type of words used.
17. The method of claim 1 , further comprising: said processor performing analysis of the chat session;
wherein said analysis considers factors that comprise any of time taken, type of chat, and recommendations provided.
18. An apparatus for chat analysis, comprising:
a processor creating and using a model for classifying resolutions in a chat session in phases comprising a model training phase, a model testing phase, and a model application phase;
a first chat filter, during said model training phase, extracting relevant portions of a chat session;
said processor performing feature extraction on said relevant portions of said chat session to obtain feature vectors;
said processor determining said feature vectors according to category relevancy;
said processor scoring said feature vectors based on a probability or likelihood of classifying said feature into a particular one of a plurality of categories;
said processor ranking said categories by their scores; and
based on a most likely category, as determined by said ranking, said processor making a final assignment of a predicted classification of said feature;
a second chat filter, during said model testing phase, said second chat filter extracting relevant portions of a chat session;
said processor analyzing said chat session; and
said processor validating said model against labeled data; and
during said model application phase, said processor receiving user chat information comprising a user interaction during a chat session across a computer network from a computer chat system or a system for real time communication across said computer network;
said processor extracting unlabeled data from said user chat information;
said processor scoring said unlabeled data with said model;
during said chat session, said processor analyzing said user interaction with said model to predict a type of chat session;
said processor determining whether said chat session comprises an information based chat query or an action request-based chat query; and
based upon said type determination, said processor determining an action to take in connection with said chat session;
when said chat session comprises an information based chat query, providing responsive information to said user; and
when said chat session comprises an action request-based chat query performing a responsive action on behalf of the user.