Context and problematic

Client : the teams in charge of supervising the activity of IT support.
Need : Have a global and precise vision of all areas to better allocate resources, reduce tickets and be pro-active on emerging/ growing themes.
Constraint : Data (tickets) not annotated and not standardized between the teams (templates, chat, emails,…)

Goals

Identification of request topics (tickets) with 3 levels of precision (ex: Outlook – Certificate – Expiration)

Identification of trends for each subject (emerging, constant, more topical)
Identification des tendances pour chaque sujet (émergent, constant, plus d’actualité)

Unsupervised analysis consistent with feedback from those responsible for each scope (difficult to quantify because unsupervised)

Our intervention

1 Data Architect for V1

• Definition / framing of the project
• Data cleaning and standardization
• Ticket analysis based on unsupervised Machine Learning methods:
  Clustering of tickets (LDA, DBSCAN, K-Means…)
  Topic Modeling / characterization of clusters
  Trend detection on identified topics
• Automation of the analysis and the final report that shall present the results

Results

V1 carried out for 2 perimeters, discussions in progress for the putting into production and the launch of a V2.

For V1:
Deliverable: automated report
Subjects identified & degree of precision: OK for the 2 perimeters
Subject trends: OK for both perimeters

Technical environment

Python
Clustering : Scikit-Learn, Gensim
Topic Modeling : Gensim, Wordcloud,
Visualisations / Restitution: Plotly, Jupyter
Jenkins, Git