What is SemDis?
SemDis is a metric that calculates the maximum semantic distance between a given word and each word in a corresponding response (read the paper). It utilizes multilingual versions of BERT or RoBERTa, depending on the language, to compute these semantic distances. SemDis is primarily designed for the Alternate Uses Task but also works well with other word association and divergent thinking tasks.
How do I use SemDis?
To calculate semantic distances using SemDis, upload a CSV file that has a column named 'item' (containing the target word from the Alternate Uses Task or similar tasks) and another column named 'response', which contains the corresponding response text. The algorithm is case-sensitive for the item/response column names, but it does not matter if other columns are present in the input CSV.
The algorithm will add two new columns to your CSV: one named 'prediction' which reflects the model's creativity score and a column named 'modelname', which is added to reflect the model providing the score.
If the language you wish to score in is not currently supported, we recommend translating the items/responses to English via Google Translate. SemDis has been found to perform well in such scenarios, according to the findings of Patterson, Merseal et al. (2023).
