We present a prototype of a novel interactive food recommender for groups of users that supports groups in planning their meals through a conversational process based on critiquing. The system comprises two novel elements: a user interface and interaction design based on tagging and critiquing, and a utility function incorporating healthiness and diet compliance factors.
Despite current technological progress, following a healthy
diet is still a challenge. While there are numerous
information resources such as books, websites, and mobile apps,
nding a suitable meal plan is not easy, mainly due to the
majority of resources being generic and non-personalized.
As such, meals suggested by a diet may not appeal for users
and they will not stick to the diet guidelines. Food
recommender systems tackle this problem by generating
personalized meal plans. They exploit food and recipe data, explicit
and implicit food preferences (e.g., ratings and browsing
behavior) to train predictive models and deliver personalized
food recommendations to inform the meal plans [
Many of the existing works on food recommendations
focused on individual users. However, in real-life scenarios the
food is consumed by a group of people, e.g., family, friends,
or school canteen. Moreover, it has been shown that small
groups of close users, such as families, have a profound role
and supportive e ect on health promotion within the groups
[
In summary, the contribution of this work is two-fold: (i) a novel interface design that applies interactive strategies, such as tagging and multi user critiquing, and (ii) a novel recommendation algorithm that generates a long term diet plan for users within the group. 2.
One of the main considerations related to food
recommendations refers to the recurrent nature of eating and food
consumption. Indeed, people eat similar times a day and
every day, and they plan their meals in a sequential manner.
Although some works focused on sequential
recommendations [
Note that this computation disregards other factors that may a ect util(u; t; m). Among these are cost(m) { the estimated cost of cooking m, avail(u; m) { the availability of the ingredients of m to u, and seq(plan; m) { the match of m to food consumption patterns observed in plan. At the current stage of our work, we focus primarily on rat, highlight diet and health as two in uential factors, and leave the remaining factors for future research.
Once the utility of m is computed for every user u 2
g in the group g, predicted score for the entire group is
quanti ed through aggregating the individual utility scores,
as per util(g; t; m) / Pu2g util(u; t; m) [
We are developing an Android prototype food recommender for groups that is depicted in Figure 1. We will brie y present the interfaces and the envisaged interactions.
In the preference elicitation phase, we ask users to specify
which meals they have ever eaten or cooked (see Figure
1left). This is done by presenting a tree of meals, e.g., pasta
! spaghetti ! spaghetti bolognese. Users can navigate the
tree and mark familiar meals. Next, users are asked to rate
and tag meals selected from: (1) familiar meals marked by
the user, (2) meals selected by an active learner [
Given the rating and tagging input, we populate the item tag matrix and incorporate the attitude into the matrix. Then, an extended matrix factorization algorithm (with content and tag features) is run to compute the predicted rating rat(u; m). The computation of diet(plan; m) and health(m), as well as of other factors a ecting util(u; t; m), depends on the constraints of the user's diet and nutritional considerations and is left beyond the scope of this paper.
Apart from the individual rating and tagging elicitation,
we also use group-based preference elicitation, in which we
di erentiate the role of the group leader, called the cook.
Based on the tags and ratings of all the group members,
the system delivers to the cook the meal recommendation
using the group utility score (Figure 1-right). The cook can
either accept the recommendation, reject it outright (e.g.,
due to unavailability of ingredients), or criticize it using the
tags (e.g., \less spicy"). Similarly, each group member can
criticize the meal accepted by the cook through a series of
interactions with the recommender. The system then
aggregates individual critiques into a compound critique by
applying a voting mechanism and resolving possible con icts,
and recommends alternative meals [
In this paper, we have presented the prototype of an interactive food recommender for groups. It features a novel utility function, a tag- and rating-based preference elicitation interface, group recommendation aggregation, and critiquebased conversational recommendations.
We are now implementing the proposed system and working on the user study design. In the future, we plan to apply a mechanism that considers the role of users in the group when aggregating recommendations. This is due to the higher impact of the preferences of dominant users (e.g., mother in the family) on the decisions of other group members (e.g., kids). We also plan to weigh di erently the predictive rating and critiques of the group members. A group member with strict diet constraints like diabetes, may can be assigned a higher importance than other members.