Social-network-graph-link-prediction

Problem statement

Given a directed social graph, the model has to predict missing links to recommend users (Link Prediction in graph)

Data Overview

Taken data from facebook's recruting challenge on kaggle https://www.kaggle.com/c/FacebookRecruiting
Data contains two columns source and destination. The source and destination columns contain information about the users and a source, destination pair signifies that they are connected in the network. Here, a link between the user u1 and u2 denotes a 1-way relationship i.e u1 follows u2 - Data columns (total 2 columns):
- source_node int64
- destination_node int64

Above is the snapshot of the raw data in train.csv file. It only has two columns source_node and destination_node. This is a pure graph based link prediction problem as we have no other meta information

Mapping the problem into supervised learning problem:

Generated training samples of good and bad links from given directed graph. For each link got some features like no of followers, is he followed back, page rank, katz score, adar index, some svd fetures of adj matrix, some weight features etc. and trained ml model based on these features to predict link.

Business objectives and constraints:

No low-latency requirement.
Probability of prediction is useful to recommend highest probability links

Performance metric for supervised learning:

F1 score
Confusion matrix

Precision: Suggesting connections that are most likely to be correct
Recall: We try and not to miss out any connections
Since we want both precision and recall to be high, we choose F1 score as a KPI along Confusion matrix to summarize the performance of the model

imamitsingh / social-network-graph-link-prediction Goto Github PK