Method for Detecting Near-duplicate Recipe Creators Based on Cooking Instructions and Food Images

In this study, we developed a method for detecting near-duplicate recipe creators based on the posting time interval and food images. Some recipe websites contain pairs of recipes with very similar cooking instructions (near-duplicate recipes), which are not useful when searching for recipes. In addition, we found that some recipe creators posted many recipes over a short period of time. Thus, we investigated whether "near-duplicate recipe creators" can be detected by using the cooking instructions and food images as clues. We evaluated our proposed method for near-duplicate recipe creator detection. In terms of the F-score, our method achieved high performance of 0.889 by detecting 19 recipe creators who posted over 30 recipes in one hour, and 0.778 for 46 recipe creators who posted over 20 recipes in one hour.