Keywords

1 Background

1.1 Underrepresented Minorities Pursuing Graduate Degrees in Computing

There is a significant lack of minority representation in graduate level computing programs and careers [1, 2], and there are many attributing theoretical factors that continue to influence this deficit. The two most substantial factors are: (1) inadequate sense of belonging and (2) insufficient self-efficacy [3, 4]. The lack of minority representation in computing departments and the rarity of minority-focused research contributes to reduced feelings of belonging in the computing arena identified by prospective underrepresented minority computer scientists [5]. Additionally, both ethnic stereotype threats and the lack of effective resources geared toward preparing underrepresented minority students for computing careers and higher education contribute to a fragile state of academic and work performance [6, 7].

1.2 Virtual Mentoring

Research suggests that effective mentorship is pivotal for student persistence and achievement, particularly for underrepresented minorities in STEM. Active mentorship helps mitigate the performance deficit of underrepresented minorities in computing by establishing an ongoing support system and guiding direction [8, 9]. Mentor relationships are effective when the interaction between mentor and mentee consists of personal connections, matching personalities, and relatability, given the mentor’s advisement and expertise is credible [10,11,12]. However, due to the lack of underrepresented minorities in the field, pairing students with more relatable mentors is challenging [13, 14]. Students accessing virtual mentoring systems viably supplements existing mentoring practices [15]. Virtual mentoring is the use of human mentors, distance technology, and computer programming to facilitate a mentoring relationship [15]. Conversational agents (or chatbots), a particular virtual mentor system of interest, are computer programs that use natural language conversations to communicate and advise human users [16]. Due to its accessibility and adaptability, conversational agents as supplemental mentoring are quite useful because they utilize different platforms such as social media and short message service (SMS), which increases accessibility. Content is interchangeable with different subjects, allowing for the advisement of students with varying interests and backgrounds, given a robust expert or intelligent design [17].

There have been very few conversational agents that have been used for advising and teaching Black computing students. Conversational agents on social media, SMS, and web-based programs have been used to discuss race and ethnic differences in health-related topics [18,19,20] as well as to provide HBCU students with general information on undergraduate prep, completion, and graduate school topics [21,22,23]. However, there has not been an intelligent system that understands its underrepresented minority users’ preferences and their background to mentor users with advisement and resource-sharing, making it to and succeeding in graduate school or a computing career. In order to build such a system, it is critical to improve the understanding of existing mentoring relationships of underrepresented minority computing students and user preferences.

2 Method

2.1 Participants

Twenty volunteers participated in the study during a conference, focusing on black and underrepresented minorities in computing. The focus groups were housed in two rooms, where focus group facilitators recruited conference attendees to volunteer to take part in the study. These volunteers were offered $40 gift cards to participate in the research. All participants were Ph.D. students or Ph.D. candidates in computing and were underrepresented minorities (Black and LatinX). Participant ages ranged from 22–54.

2.2 Procedure

The lead authors facilitated the two focus groups, and a semi-structured model was used for each group. Each focus group lasted one hour to discuss how African-American computing students view mentors and their recommendations for developing an effective virtual mentoring system. The participants were asked to describe their current mentors and how they interact and deliver advisement, as well as how they reacted to being mentored and if they were satisfied with their mentoring relationships. To follow, participants were asked about their previous experiences with virtual mentoring systems and how they would develop an intuitive, useful tool to reflect a supplementary mentor. The focus group was audio recorded and later transcribed for data analysis.

2.3 Analysis

A direct hybrid inductive-deductive thematic analysis was employed to examine the transcriptions from the focus groups [24]. A code manual was developed by synthesizing literature suggestions on significant factors for mentoring underrepresented minority students in computing, and the factors that contribute to developing effective virtual mentoring systems [11]. Transcribed responses were inductively summarized into initial themes prior to being categorized by the codes from the code manual. Common themes were synthesized and theme definitions were revised for further legitimization.

3 Results

3.1 Current Mentor Description

The participants reported that their current mentors were more knowledgeable than them. According to one participant, a mentor is “Someone who’s a little bit more senior than you, who’s been in the business, who knows how to code in Java or knows the system a little better than you and basically that provides this day-to-day interaction.” In many instances, a mentor is assigned to the mentee at work without consideration of their social or personal background, “being a new coder on entry level, they always assign you a mentor.” In agreement, all participants expressed how mentors are great for making connections that could advance their professional careers. Participants also described themselves having multiple mentors for different aspects of their lives, “I feel that a lot of times when I’m in different positions, I try to find someone who I can connect with”.

3.2 Mentor Communication Delivery

The responses for mentor advice delivery ranged from very positive interactions to very negative interactions. Some participants recommended having more than one mentor so “depend[ing] on how you feel”, you can cater the kind of advice you receive. For instance, one participant stated that sometimes they “need to tough love. So [they] go with somebody who can give [them] tough love. Or ...[they] feel ...very down and [they] need someone to encourage [them].” One participant described their experience as a respectful one, “never condescending.” However, on the opposite end of the spectrum, another participant described their experience with advice delivery as negative, and stated that the advisor was “...very direct ...it can come across condescending, or it can come across as not necessarily encouraging.” In a similar sense, many participants felt their advisors were straightforward when delivering advice.

3.3 Mentee Reaction

Overall, the mentee reaction to human advisement was positive. One mentee highlighted the importance of person-to-person interaction when they exclaimed, “...you are getting the advice, or the response from like a real person.” It was reiterated that having multiple mentors is most helpful because “...whenever there’s a problem, or situation, or concern, or want feedback on something, [they] get multiple perspectives every time.” Claims have also been made that the advisement process has helped some mentees with becoming less sensitive to constructive criticism. One of the participants stated that they “...used to take stuff personal, like, when [they] first started [their] Ph.D. program ...”, which is when they met their mentor.

3.4 Mentee Satisfaction of Advisement

Some participants believed having a mentor that they could identify with is important to forming a strong relationship. One participant attended “an all black, male HBCU”, so in his opinion, he was “fortunate enough to have mentors that looked exactly like me, black males”; on the contrary, some participants valued having a mentor that experienced situations similar to theirs, however they found it challenging to make a connection with one. Others felt that it was acceptable to not have a mentor that looked like them. Mentoring relationships stem from a mentor wanting to help guide or advise someone: “Everyone wants a mentor that will pull them up through the ranks, but not every mentor will see something worth cultivating in every individual that approaches them, which makes whomever they choose to enter that relationship unique.”

3.5 Experience with Virtual Mentor Systems

The participants expressed little to no experience with using a virtual mentoring system. One of the participants reported: “I have zero experience with it. But one of my Co-PI, he was showing me this site. Basically it’s kind of like an outsourcing agent.” As a whole, the participants expressed that virtual mentors cannot be used as a blanket for multiple subjects, or for in depth questions: “it’s not like a full mentor, but it’s just if you need help right now, type in your question or can you give me advice instantly? And it’s something automated ...It’s a good tool for certain instances. But I don’t think it’s a coverall.”

3.6 Virtual Mentor System Recommendation

We received a large amount of feedback regarding the virtual mentor system. The most prominent recommendation was to develop a more personal relationship. For instance, one participant stated that trust was an important part of mentorship and that the system lacks “rapport” in addition to having those things that you build from, like “interactions.” Another noteworthy recommendation was having a larger range of predetermined information. A participant stated that “...it’s hard to have a dynamic conversation with a virtual mentor unless it’s a real person”, so in order for the system to be most efficient and to understand exactly what [users are] looking for, have “...as much predetermined information as possible for any wide range of topics, that may come up.”

4 Discussion

Many participants were engaged in ongoing mentoring relationships; mentorships are suggested as beneficial and sometimes necessary for successful pursuit of computing careers [8, 9]. Mentors have teachable skill sets and work experience that aided their advice credibility [11]. The mentors’ age ranged; a few participants had a near-peer mentor who had recently obtained their Ph.D. Some participants did not have what they considered a mentor, but all agreed that a mentor whom you could connect with was essential for a healthy mentoring relationship [12]; thus the idea of mentors with welcoming and friendly personalities and approaches was highlighted. To accommodate, the mentor system will require a very clear description of its scope and responses with a conversational flow that is highly intuitive. Few participants had assigned mentors from work, research experiences, or academic programs, but all mentoring relationships formed from these experiences. Given this, it will be useful for the intelligent conversational agent to be introduced through a work, research, or academic experience. Advisement was delivered in a considerate, constructive, relatable, and straight-forward approach [12, 14, 15]. As a relatable system, serious planning needs to occur to ensure users feel the system is speaking to them as a human on an individual personality level, as well as with a sociocultural lens. In addition, further research must be done to understand what considerations are necessary to fine-tune the information needed to be obtained from users. One participant described how their mentor used humor: “being able to share a joke with them afterwards. Or during. Because that’s how I receive information. I receive information, that’s when I feel like we’re having a conversation that matters. But it doesn’t have to be so constricting that I have to feel like we can’t talk like human beings.” A humor feature would need to be further researched and developed to determine its benefits, appropriateness, and suggested dosage to make the feature effective.

Participants appreciated their mentors’ advice, encouraging them to think critically and were even used to receiving information that they may not have expected, as it motivates and humbles them [10, 15]. Mentees also enjoyed feeling their mentor cared and showed humility and honesty [10, 15]. An intelligent mentoring system must not only be accurate, but also deliver the message in a way that makes the user feel like the creators want them to succeed and modestly may not always have the best advice for them. Participants discussed how their satisfaction in a mentoring relationship is also influenced on the overall availability of the mentor. As a primary rationale for the system [18], an intelligent virtual agent would need to be accessible on many platforms and devices at as much time and in as many locations as possible. It is expected that the system would be accessible anytime at any position that supports a user’s service on the varying platforms and the devices that possess it. One participant discussed how they liked that their mentor visually and socially reflected them [7, 9]. Having a system be reflective of the user is very important and can be approached in many different ways, such as using location based subjects, examples, terminology, and allowing users to choose a mentor avatar that they feel most comfortable with. Mentees did not take joy in their mentors having a disdainful tone and being unfair or unreasonable. Being careful in the word choice of responses based on the subject matter, flow, and timing of the conversation is essential to maintaining a supportive user experience.

Quite a few participants never used a virtual mentor, however there are those who have used supplemental video conferencing and screen sharing with a person, personal assistants like Siri, and automated chatbots. Their experiences with them varied. Participants claimed their interactions were a waste of time, describing their system as uncommitted and untrustworthy with its responses and ability to retrieve classified information. Others believed it was interesting and helpful, though not always necessary. Confidentiality must be explicitly stated, describing proof of privacy for some users to trust a developed mentoring system, and there were many recommendations for developing this system. Being highly secure and having individual sign-in access helps to maintain the sense of privacy and confidentiality as well. Having evidence-based responses and direct resources from the internet is essential to resource-sharing capability, and being weary of bias and stereotyping is critical for the relatability and demographic target of the tool. The system intends to be mapped with SMS and social media, which will be automatic, and establishing a website portal is very useful as well. Integrating through social media allows for more user information to be collected to better tailor intelligent responses. Voice activation and functionality helps to expand the reach of the tool as well as the usability. Having an image to accompany voice and messaging is essential to allow users to feel as if they are connected with a person rather than a computer. Other interesting and more reaching suggestions included having the conversational agent have the same personality as the mentee, integrating users’ music playlists, and having random social interaction.

There were few limitations in this study, including that the sample only contained Ph.D. students and candidates. This is great for giving recommendations and sustained mentoring experiences with the profession of computing, however this is limited as it does not address undergraduate students, a likely user demographic. It also does not include underrepresented minorities working in industry. Furthermore, the sample was small, consisting of a total of 20 participants amongst the focus groups. Though the codebook was developed through validated literature suggestions, an existing validated codebook was not used.

5 Conclusion

This study explored the mentoring relationship of underrepresented minority computing students to provide a baseline for design considerations for an intelligent conversational agent to mentor those who would like to pursue a career in computing or enter and complete a graduate program. This study also gauged underrepresented minority computing students’ experiences and recommendations for virtual mentoring systems. Findings will be used to develop an intelligent mentoring conversational agent for underrepresented minority computing undergraduate and graduate students. Findings are also useful for mentoring and computing outreach professionals interested in supporting underrepresented minority computing students.