Homophily, the tendency for individuals to associate with others who share similar attributes, is a fundamental concept in network analysis and has recently been extended to higher-order networks such as hypergraphs. Existing methods quantify higher-order homophily based on categorical node attributes, but they are limited to static networkss and do not address temporal hypergraphs where node categories change over time. In this paper, we propose a method for analyzing higher-order homophily in temporal hypergraphs, both for individual categories and for category pairs. We apply the proposed method to a temporal hypergraph of ingredients constructed from a Japanese recipe-sharing website, where weekly varying ingredient categories are derived using a burst detection technique, and demonstrate their effectiveness in uncovering category-level homophily patterns, offering new insights into the co-occurrence patterns of ingredients in Japanese homemade cooking.