Description
Draw networks of entities mentioned in the same sentence, cluster dense groups, and step through time to watch links appear and vanish, drilling down to source sentences.
Entities and categories are first extracted from a text corpus. Each becomes a node, and an edge links two entities that appear in the same sentence, weighted by frequency. Category connection maps place chosen categories around a circle; relationship maps add a context node; spring layouts handle larger sets. Dense groups are found by iterating a neighbour-based weight (Feldman uses five iterations), keeping the top 10% of edges and taking connected components. Trend graphs add a time slider so the analyst can see links strengthen, appear or disappear across periods. Every edge can be opened to the documents and sentences behind it, which is what keeps the method honest: co-occurrence is only a candidate relationship until the text confirms it.
Attributes
- Discipline
- Qualitative Analysis & Synthesis
- Loss type
- Unmigrated
- Stage
- Discovery, Post-launch
- Data
- Quantitative
- Effort
- Days
Steps
- Extract entities and categories (products, features, organisations, issues) from the corpus.
- Count same-sentence co-occurrences for every entity pair.
- Choose categories of interest and draw a connection map or spring layout.
- Filter weak edges (for example keep the top 10%) and cluster dense subgraphs.
- Step through time windows to see which links appear, strengthen or disappear.
- Open the supporting sentences for each important edge and confirm the relationship before reporting it.
Sources
- Handbook of Data Mining (Ye, 2003)
Also known as
- Entity co-occurrence maps
- Category connection map
- Relationship map
- Co-occurrence network
- Trend graph
Categories
- User Research
- Qualitative Analysis & Synthesis