Towards Contextual Healthiness Classification of Food Items - A Linguistic Approach

International Joint Conference on Natural Language Processing, pages 19–27, Nagoya, Japan, 14-18 October 2013. Towards Contextual Healthiness Classification of Food Items - A Linguistic Approach Michael Wiegand and Dietrich Klakow Spoken Language Systems Saarland University D-66123 Saarbr¨ucken, Germany {Michael.Wiegand|Dietrich.Klakow}@lsv.uni-saarland.de Abstract We explore the feasibility of contextual healthiness classification of food items. We present a detailed analysis of the lin- guistic phenomena that need to be taken into consideration for this task based on a specially annotated corpus extracted from web forum entries. For automatic classi- fication, we compare a supervised classi- fier and rule-based classification. Beyond linguistically motivated features that in- clude sentiment information we also con- sider the prior healthiness of food items. 1 Introduction Food plays a substantial part in each of our lives. With the growing health awareness in many parts of the population, there is consequently a high de- mand for the knowledge about healthiness of food. In view of the variety of both different types of food and nutritional aspects it does not come as a surprise that there is no comprehensive repository of that knowledge. Since, however, much of this information is preserved in natural language text, we assume that it is possible to acquire some of this knowledge automatically with the help of nat- ural language processing (NLP). In this paper, we take a first step towards this endeavour. We try to identify mentions that a food item is healthy (1) or unhealthy (2). (1) There is not a healthy diet without a lot of fruits, vegetables and salads. (2) The day already began unhealthy: I had a piece of cake for breakfast. This task is a pre-requisite of more complex tasks, such as finding food items that are suitable for cer- tain groups of people with a particular health con- dition (3) or identifying reasons for the healthiness or unhealthiness of particular food items (4). (3) Vegetables are healthy, in particular, if you suffer from diabetes. (4) Potatoes are healthy since they are actually low in calories. The major problem of identifying some Is- Healthy or Is-Unhealthy relation is that the simple co-occurrence of a food item and the word healthy or unhealthy is not sufficiently predictive as shown in (5)-(7). (5) Chocolate is not healthy. (6) The industry says chocolate is healthy, but I guess this is just a market- ing strategy. (7) If chocolate is healthy, then I will run for the next presidential election. We describe the contextual phenomena that un- derlie these cases and provide detailed statistics as to how often they occur in a typical text collec- tion. From this analysis we derive features to be incorporated into a classifier. Our experiments are carried out on German data. We believe, however, that our findings carry over to other languages since the aspects ad- dressed in this work are (mostly) language univer- sal. For the sake of general accessibility, all exam- ples will be given as English translations. To the best of our knowledge, this is the first …

Content truncated to 500 words