# string clustering: similarity criterion

I have a set of strings of dimension $10,000$. I want to group similar strings together in one group, perform clustering. As string metric, I am using the Levenshtein distance.

Simply, with the Levenshtein distance I'll just compute distance between $2$ strings and then by using a threshold the clustering algorithm will make the decision if they can be grouped or not. This is not enough. I am looking for a special measure to study the relation between the strings.

For example: door and entrance wont be grouped together if I just compute the Levenshtein distance, in fact there is nothing in common between these $2$ words. But logically they are connected and can be grouped together, since door and entrance are basically the same.

• Have you ever come across such a problem?
• Is it similar to the Semantic similarity measure?

## 1 Answer

Have a look at WordNet. Quoting from http://wordnet.princeton.edu/:

WordNet is a large lexical database of English. Nouns, verbs, adjectives and adverbs are grouped into sets of cognitive synonyms (synsets), each expressing a distinct concept. Synsets are interlinked by means of conceptual-semantic and lexical relations. The resulting network of meaningfully related words and concepts can be navigated with the browser. WordNet is also freely and publicly available for download.

• yes I was just looking @micans. I found the wordnet website too wordnet.princeton.edu/wordnet/download/current-version I wanted to make sure that I am going in the right direction. What I am looking for is semantic similarity right? Mar 11, 2014 at 11:27
• Semantic similarity does not seem to be a very well defined technical concept, more a loose general term, so I would not worry about 'doing it correctly'. The most important thing is what you want to get from your data. Certainly, WordNet is a highly useful resource and will help you with meaning of words. It is a good place to start. Regarding 'Semantic similarity' (e.g. the wikipedia page), I would just use it to gain some background knowledge and/or ideas for future reference. The crucial aspect to all this is your data and what you hope to get from it though. Mar 11, 2014 at 11:40