How can you leverage different types of explanations and modalities for explainable recommender systems?

1 Types of explanations

Explanations can take many forms, depending on their content, format, and purpose. Feature-based explanations emphasize the attributes or characteristics of the recommended items or the user profile that influenced the recommendation. Rule-based explanations provide the logic or criteria behind the recommendation. Case-based explanations show examples or analogies of similar or relevant items or users that support the recommendation. Social explanations leverage the opinions, ratings, or feedback of other users or experts to endorse the recommendation. And causal explanations reveal the causal or counterfactual effects of the recommendation.

2 Modalities of explanations

Explanations can be presented to the user in different modalities. Textual explanations use natural language and can range from a sentence to a story. Visual explanations use graphical elements, such as icons, images, charts, or diagrams, and can be static or dynamic. Auditory explanations communicate information with sound, such as speech, music, or effects, and can be verbal or non-verbal. Multimodal explanations combine two or more modalities to provide an enhanced experience and can be synchronous or asynchronous.

3 How to leverage explanations and modalities

Explainable recommender systems can be leveraged in different ways, depending on the user's goal, preference, and context. For instance, if the user wants to explore new options, feature- or case-based explanations with visual or multimodal modalities can be used. On the other hand, if the user wants to verify the quality of the recommendation, rule-based or social explanations with textual or auditory modalities can be provided. It's important to balance the complexity and effectiveness of the explanation; too simple an explanation may not provide enough information or credibility, while too complex an explanation may overwhelm or confuse the user. You can adjust the complexity of the explanation by using different levels of detail, transparency, and personalization. Finally, you should evaluate the impact of the explanation and modality on the user's behavior, satisfaction and trust. This can be done by measuring user engagement, feedback and retention with the recommender system as well as conducting user studies, surveys or interviews to collect qualitative data on their perception and experience.

4 Examples of explainable recommender systems

To illustrate the range of explanations and modalities that can be used for explainable recommender systems, here are some examples. MovieLens offers feature-based and social explanations with textual and visual modalities, such as predicted ratings, average ratings, number of ratings, genres, tags and actors of the movie. Yelp is another well-known online platform that provides rule-based and social explanations with textual and visual modalities such as user location, preferences and search history. Xplain is a prototype of a multimodal explainable recommender system that uses speech and gestures to interact with the user. It provides feature-based and case-based explanations with auditory and visual modalities, such as natural language generation and projection mapping to display the explanations on physical objects.

5 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?