IBM Research-Ireland

Harnessing Dialogue for Interactive Career Goal Recommendations

Share this post:

Career recommendations represent a unique opportunity to engage users with an interactive recommender. In our ACM IUI Conference paper “Where Can My Career Take Me? Harnessing Dialogue for Interactive Career Goal Recommendations”, we present an interactive career goal recommender framework that leverages the power of dialogue to allow users to interactively improve the recommendations and bring their own preferences to the system.

Career goals represent a special case for recommender systems, as they require considering both short- and long-term goals, which introduces challenges that standard techniques cannot handle.

Existing recommendation algorithms focus on the immediate value of recommendations, taking into account the current needs of the user. Career goals, on the other hand, are not just about looking for the next thing, but about helping users identify long-term goals and letting them explore where their careers can take them. At the same time, the goals do need to be achievable and relevant to users, so there is a trade-off between relevance, achievability, and aspiration.

Figure 1: Eliciting User Motivation

People may have different motivations and concerns when considering new long-term goals, so involving them in the recommendation process is important. For example, consider two people with similar job histories and skillsets. One of them has a clear career goal in mind and is happy with her current path, whereas the other might feel her current role isn’t leading where she wants to be and is looking for a change. These two individuals would like to leverage their current skills while moving in totally different directions. Most recommender systems would have trouble making different recommendations for these two very different users without understanding their motivation.
Hence it is important to involve users in the recommendation generation process, so they may guide the recommender. Placing users “in the loop” ultimately increases the trust and the transparency of the whole solution. Additionally, the stakes are higher for people when making a career decision than when choosing a movie or a song to entertain them. As a result, we feel career recommendations represent a unique opportunity to engage users in an interactive recommender as we believe their motivation will be greater.

To achieve this, we developed an interactive career goal recommender (ICGR) framework that leverages the power of dialogue to allow users to interactively improve the recommendations and bring their own preferences to the system [1]. The underlying recommendation algorithm is a novel solution that suggests both short- and long-term goals through utilizing the sequential patterns extracted from career trajectories that are enhanced with features of the supporting user profiles.

Figure 2: Explore Item

Typical recommender systems present a list of recommendations, and user interaction is limited to accepting or rejecting the items. To truly engage users, we support multiple interactions to enable them to understand the recommended items and provide their feedback so that the system can generate suggestions that better suit their needs in the following conversation turns or sessions. The possible interactions allowed include asking for further details on each recommendation item, giving feedback at the item level,  and giving feedback on individual item features, such as telling which skills they would like to develop.

Users’ preference information will be stored in their profiles and used later on by the recommender to suggest jobs that require those preferred skills. Unlike most recommender applications, ours does not have rating information when looking at a user’s skills or job history. For example, a user’s job history might include jobs that the user did not enjoy, with no such information recorded. Interacting with the user to elicit such preferences and details becomes vital in creating a viable solution.

We address this in two ways. Firstly, we allow users to explore items and see the explanations for why items are recommended. In this way users can understand why a career goal is recommended and provide feedback if they observe any incorrect assumption the system has made regarding their interests. For example, an explanation for recommending a Senior Test Architect role could be that, “75% of people who were QA Engineers go on to be Senior Test Architects”. The role of QA Engineer may in fact be a part of a user’s profile, but if he or she potentially did not enjoy the role, the user can tell this via dialogue to the recommender and the recommender discounts the influence of the role on future recommendations.

Figure 3: Explanations and Correct Assumptions

The other mechanism we employ is to prompt the user with an explicit preference elicitation question chosen to have the highest impact on fine-tuning the current recommendations. This is performed using the Information Gain metric calculated for each skill that exists in the recommended goals. The elicitation question is formulated using the skill that has highest information gain with respect to the recommended items; therefore, once a user answers this question through selecting a skill, the recommended items will be filtered such that the ones with the selected/preferred skills will be ranked higher and the ones that involves disliked skills will be removed from the resulting list.

We performed extensive experiments with two real world datasets with typical metrics from two related areas of research: recommendation systems and dialogue systems. In addition, we designed an evaluation based on simulating user interaction with the system through dialogue. Overall, the experiment results prove that ICGR achieves better performance considering the effectiveness of the dialogues, the response rate, and the accuracy of the recommendation suggestions, as we allow more complex interactions with the users. The proposed solution is added as a new feature to IBM Watson Career Coach, which is being made available to pilot with selected users. This new functionality is also seen as a key feature for customers such as Citizens Financial Group [2].

[1] Oznur Alkan, Elizabeth M. Daly, Adi Botea, Abel N. Valente, Pablo Pedemonte. “Where can my career take me? Harnessing Dialogue for Interactive Career Goal Recommendations”. Accepted for Publication in ACM IUI 2019.

[2] https://www.americanbanker.com/news/citizens-recruits-a-career-coach-from-ibms-ai-lab

IBM Research

More IBM Research-Ireland stories

Text2Scene: Generating Compositional Scenes from Textual Descriptions

At CVPR 2019, IBM researchers introduce techniques to interpret visually descriptive language to generate compositional scene representations from textual descriptions.

Continue reading

Overcoming Challenges In Automated Image Captioning

At CVPR 2019, IBM researchers introduce an improved method to bridge the semantic gap between visual scenes and language to produce diverse, creative and human-like captions.

Continue reading

Unveiling Analog Memory-based Technologies to Advance AI at VLSI

At the 2019 VLSI, IBM researchers will present three papers that provide novel solutions to AI computing based on analog devices.

Continue reading