Machine Learning-Based Search for Similar Unstructured Text Entries Using Only a Few Positive Samples with Ranking by Similarity

This research was inspired by the procedures that are used by human bibliographic searchers: Given some textual, only 'positive' (interesting) examples, coming from one category find the most similar ones that belong to a relevant topic. The problem of categorization of unlabeled relevant and irrelevant textual documents is here solved by using a small subset of relevant available patterns labeled manually. Unlabeled text items are compared with such labeled patterns. The unlabeled samples are then ranked according their degree of similarity with the patterns. At the top of the rank, there are the most similar (relevant) items. Entries receding from the rank top represent less and less similar entries. This simple method, aimed at processing large volumes of text entries, provides practically acceptable filtering results from the accuracy point of view and users can avoid the demanding task of labeling too many training examples to be able to apply a chosen classifier. The ranking-based approach provides results that can be further used for the following text-item processing where the number of irrelevant items is already not so high as it is usually typical for, for example, only the raw browsing results provided by Internet search engines. Even if this relatively simple automatic search is not errorless, it can help process particularly very large textual unstructured data volumes. Such an approach can help also in the economics area, for example, to automatically categorize written opinions of customers (as amazon.com is collecting via the Internet), process network-based discussion groups, and so like.