Prototype
KNN Search
What if…
the search for news articles which is output by an algorithm had a mathematically comprehensible foundation?
Problem
The classifications and recommendations of algorithms do not always seem to make sense. They often cause incomprehension due to lack of context and traceability. Many outputs also turn out to be one-dimensional and rarely convince with additional value. How could similarities between two parameters - in this case news articles - be precisely determined in order to generate a real benefit?
Solution
KNN (K-Nearest-Neighbor algorithm) addresses the problem of inaccurate comparisons and classifications by dynamically calculating similarities. Based on the "lazy learning" method, queries are processed in real time. The proximity of two independent elements is calculated in order to derive conclusions about the properties of the other from the properties of one parameter. Whether the factors to be compared are quantities, coordinates, articles or ratings is the same. All calculations are based on a corresponding mathematical formula (correlation efficient according to Pearson), which makes the output of the KNN comprehensible.
