Empowering Your Business with Pure Cypher

In the previous chapters, we introduced the concepts behind Neo4j and learned how to query it using Cypher. It is now time to build our first working application of graph databases. The first step when entering the graph database ecosystem is usually to try and build a knowledge graph of your business or industry. In this chapter, you will learn what a knowledge graph is and how to build one from structured or unstructured data. We will use some Natural Language Processing (NLP) techniques and query existing knowledge graphs such as Wikidata. We will then focus on two possible applications of knowledge graphs in the real world: graph-based search and recommendations.

The following topics will be covered in this chapter:

• Knowledge graphs
• Graph-based search
• Recommendation engine

Technical requirements

The required technologies and installations for this chapter are as follows:

• Neo4j 3.5
• Plugins:
• APOC
• GraphAware NLP library: https://github.com/graphaware/neo4j-nlp
• Some minor parts of this chapter also require Python3 to be installed; we will use the spaCy package for NLP.
• GitHub repository for this chapter: https://github.com/PacktPublishing/Hands-On-Graph-Analytics-with-Neo4j/tree/master/ch3

Knowledge graphs

If you have followed Neo4j news for the last few years, you have probably heard a lot about knowledge graphs. But it is not always clear what they are. Unfortunately, there is no universal definition of a knowledge graph, but let's try to understand which concepts are hidden behind these two words.

Attempting a definition of knowledge graphs

Modern applications produce petabytes of data every day. As an example, during the year 2019, every minute, the number of Google searches has been estimated to be more than 4.4 billion. During the same amount of time, 180 billion emails, and more than 500,000 tweets are sent, while the number of videos watched on YouTube is about 4.5 billion. Organizing this data and transforming it into knowledge is a real challenge.

Knowledge graphs try to address this challenge by storing the following in the same data structure:

• Entities related to a specific field, such as users or products
• Relationships between entities, for instance,...

Graph-based search

Graph-based search emerged in 2012, when Google announced its new graph-based search algorithm. It promised more accurate search results, that were closer to a human response to a human question than before. In this section, we are going to talk about the different search methods to understand how graph-based search can be a big improvement for a search engine. We will then discuss the different ways to implement a graph-based search using Neo4j and machine learning.

Search methods

Several search methods have been used since search engines exist in web applications. We can, for instance, think of tags assigned to a blog article that help in classifying the articles and allow to search for articles with a given tag. This method is also used when you assign keywords to a given document. This method is quite simple to implement, but is also very limited: what if you forget an important keyword?

Fortunately, one can also use full-text search, which consists of matching...

Recommendation engine

Recommendations are now unavoidable if you work for an e-commerce website. But e-commerce is not the only use case for recommendations. You can also receive recommendations for people you may want to follow on Twitter, meetups you may attend, or repositories you might like knowing about. Knowledge graphs are a good approach to generate those recommendations.

In this section, we are going to use our GitHub graph to recommend to users new repositories they are likely to contribute to or follow. We will explore several possibilities, split into two cases: either your graph contains some social information (users can like or follow each other) or it doesn't. We'll start from the case where you do not have access to any social data since it is the most common one.

Product similarity recommendations

Recommending products, whether we are talking about movies, gardening tools, or meetups, share some common patterns. Here are some common-sense assertions that can...

Summary

This chapter described in detail how to create a knowledge graph, either using already structured data, such as an API result, or an existing knowledge graph that can be queried, such as Wikidata. We also learned how to use NLP and named entity recognition in order to extract information from unstructured data, such as a human-written text, and turn this information into a structured graph. We have also learned about two important applications of knowledge graphs: graph-based search, the method used by Google to provide even more accurate results to the users, and recommendations, which are a mandatory step for e-commerce today.

All of this was done with Cypher, extended by some plugins such as APOC or the NLP GraphAware plugin. In the rest of this book, we will make extensive use of another very important library when dealing with graph analytics: the Neo4j Graph algorithms library. The next chapter will introduce it and give application examples in the context of the shortest...

Questions

Using Wikidata, what kind of contextual information can we add to the repository language?

Further reading

• If you are new to NLP and want to learn more about it, you can start with Natural Language Processing Fundamentals, D. Gunning and S. Ghosh, Packt Publishing.

• Then, consider checking out Hands-On Natural Language Processing with Python, R. Shanmugamani and R. Arumugam, Packt Publishing.
• W3C specifications:
• RDF: https://www.w3.org/TR/rdf-concepts/
• SPARQL: https://www.w3.org/TR/rdf-sparql-query/
• Neural Machine Translation models used by Google Translate:
• The initial paper by Google: https://research.google/pubs/pub45610/
• The implementation using Tensorflow: https://github.com/tensorflow/nmt