Co-occurrence network

As explained in Graph Algorithms for Data Science

A hashtag co-occurrence network is a graph whose nodes are hashtags and whose relationships represent hashtags occurring together in observed content.

A co-occurrence network turns entities found together in content into a graph. In a hashtag network, each Tag is a node and a connection means two hashtags appeared in the same tweet. Connections can reveal useful tag combinations, topic clusters, and less obvious hashtags beyond the most frequent ones. section 7.0

Figure 7.5 Constructing a co-occurrence network of tags based on whether they co-occur in the same tweet

This construction view makes the projection clear: tweets link to their hashtags first, then hashtags that share a tweet become connected. section 7.0

Build a hashtag co-occurrence network in practical steps:

• Extract hashtags from original tweets, excluding retweets so repeated content does not skew counts. Normalize tags, create Tag nodes, and connect each Tweet to its tags with HAS_TAG relationships. section 7.1
• Match two Tag nodes attached to the same tweet. Keep one ordering of each pair, such as id(h1) < id(h2), to avoid returning the same pair twice. section 7.2
• Use the number of shared tweets as a relationship weight. Treat the conceptual connection as undirected; in GDS, represent it with opposing directed relationships when traversal in both directions is needed. section 7.2
• Alternatively, infer relationships with node similarity. Jaccard similarity compares the tweets associated with two hashtags, while similarityCutoff and topK control which relationships remain. section 7.2.2

Reading a weighted hashtag pair

The pair #ai and #nlp co-occurs in 1,507 tweets in the sample, so its connection can carry a weight of 1,507. High counts identify commonly paired tags, but a similarity-based projection can instead emphasize overlap relative to each tag’s broader tweet set. section 7.2 section 7.2.1

Network density shapes the interpretation. Raising topK gives each hashtag more similar neighbors, while raising similarityCutoff removes weaker links; broader networks can produce broader communities. Analyze the resulting Tag-only network with degree centrality, label propagation, and PageRank to inspect connected tags, communities, and community representatives. section 7.2.2 section 7.4 section 7.5

How Graph Algorithms for Data Science covers it

How deep it goes

Walks through extraction, Cypher construction, weighting, similarity-based inference, density tuning, and community analysis.

Examples

Uses tweets and hashtags, including #ai–#nlp pairs, alongside keyword, character, ingredient, and product-basket networks.

What it assumes

Teaches elementary hashtag extraction and Cypher list handling from stored Tweet text.

What you can do after

Apply it via Cypher and Neo4j Graph Data Science to infer hashtag networks and inspect their communities.

How it connects

Connects bipartite networks, Jaccard similarity, node similarity, degree centrality, label propagation, and PageRank.

liveBookmanning.com

Read Graph Algorithms for Data Science on liveBook Get the book at manning.com

Related concepts

• weakly connected componentused by · A weakly connected component is a maximal set of vertices in a directed graph that remains connected when edge directions are ignored.
• bipartite graphproduced by · A bipartite graph is a graph whose vertices can be partitioned into two disjoint sets, with edges only between the sets.
• Cypher projectionproduced by · Cypher projection is a graph projection technique that uses Cypher queries to select, filter, and transform a graph subgraph before loading it into memory.
• co-occurrenceuses · Co-occurrence is the presence of two or more items, events, or attributes within the same observation, context, or defined sequence, without necessarily implying causation.
• node similarity algorithmproduced by · A node similarity algorithm computes similarity scores or relationships between graph nodes from shared or comparable structural features.
• label propagationused by · Label propagation is a graph-learning technique that iteratively propagates node labels through connections to assign labels or identify communities.

This explainer was generated by Manning's knowledge graph from the text of Graph Algorithms for Data Science. The section links point to the parts of the book it draws on.