A free fortnightly newsletter on Taxonomy, Thesauri & Ontology and Semantic Publishing
Deep Link Graph Analytics is Powering the Next Advance in Machine Learning
Applying graph database capabilities to machine learning (ML) and artificial intelligence (AI) applications is a relatively new concept. This is surprising considering Google’s Knowledge Graph dates back to 2012. Also graphs are a natural fit, as they are ideal for storing, connecting, and making inferences from complex data. In this article, the authors deep dive into how graphs can help ML and how they are related to deep link graph analytics for big data.
Graphs perform a significant role in unsupervised native graph-based machine learning algorithms, training supervised machine learning algorithms, explainable ML/AI models and in native parallel graph databases.
First, graph analytics offer a unique set of unsupervised ML methods. For instance, some hosts of graph algorithms can identify meaningful graph-oriented patterns, which have wide applications. These graph algorithms need to conduct deep link graph analytics, which requires superior computational capabilities offered by a native parallel graph database.
Another way deep link graph analytics help is by enriching the set of data features available for supervised ML. Consider the example of China Mobile. The company leveraged graph-based ML features like stable group and in-group connections and transformed its fraud detection process. This helped China mobile generate 118 graph-based features for each phone, thereby feeding billions of new training data record to their ML solution.
Additionally, graph analytics assist in ML and AI by helping to extract human understandable insights from them. For example, explainable models are used to highlight the key variables that lead to a decision in ML. A traditional example of an explainable model is a decision tree, which is nothing but a specialized graph. Similarly, when graph algorithms or graph features are used as part of an AI model, the natural semantics of graph relationships, such as “Customer — (bought) –> Product” lend themselves easily to interpretation. In addition, graph analytics are well-suited to compute and show the evidence behind personalized recommendations and explain the recommendation using graph visualization.
Furthermore, graph-based ML and analytics are helpful to enterprises. For instance, in fraud detection, it enables investigators to see, visually and interactively, how a transaction in question is connected to those previously marked as fraudulent.
Finally, there is an indispensable need for a native graph database featuring massively parallel and distributed processing capabilities. This can accommodate ML’s peak computational demand. For example, to compute and explain the reasons behind personalized recommendations and fraud detection, the graph database needs a powerful query language that can traverse the connections in the graph. In addition, it needs to support computation such as filtering, aggregation, and complex data structures to remember the evidence. This is where a native graph database with their parallel and distributed processing capabilities can help.
In short, deep link graph analytics is powering the next advance in ML, through unsupervised learning of graph patterns, feature enrichment for supervised learning, and providing explainable models and results. Combined with AI and ML, deep link graph analytics will serve enterprises well for years to come.
Click here to read the article.