[probably_wrong]

If you think two items cover the same highly similar _keywords_ then Jaccard distance should work.

If you think two items share highly similar _text_ then give Levenshtein distance a try.