The overall objective of this study was to identify suitable combinations of automated text processing techniques, which investigators call “Non-Intrusive Psycho-Social State Assessors” or NIPSSAs and data streams for assessing psychosocial states of interest to NASA and to validate their overall hypothesis that NIPSSA measures could replace survey data for assessing selected psychosocial states. Astronauts’ journals, diaries, and blogs are a rich source of material about their experiences and attitudes toward those experiences, and interpersonal communications are, after all, one of the best methods humans have of assessing changes in attitudes and states. The investigators reasoned that if they could automate assessments of such data streams, many of which are already being captured and, to some extent, monitored during space missions, they would greatly expand the set of data which NASA team dynamics researchers have available.
The third project year was devoted to completing the bed rest analog journal study, while continuing to participate in a four-month crew habitat study in the Hawaii Space Exploration Analog and Simulation (HI-SEAS) and began participation in an eight-month HI-SEAS study collecting both journal and crew-ground textual interaction messages (with corresponding survey data). Finally, techniques for audio data collection and then transcriptions were developed in the Human Exploration Research Analog (HERA), from which investigators obtained and analyzed journal, survey, and text and speech interaction data from four separate week-long missions.
In the fourth and final year, investigators completed analyses of the HERA transcription data, and performed cross-analog analyses illustrating the robustness of their techniques. In the HERA study, they collected more than 160,000 minutes of captured audio recordings across four mission and transcribed large portions of this to enable examination of automated processing of interactive speech data. In all three analogs, investigators collected participant journals and daily surveys to cross validate their automated assessment techniques.
The results show that automated textual analysis of subjects’ free form journals can reliably replace at least some forms of attitude and emotion surveys since the automated journal analyses regularly correlated significantly with subjects’ own survey ratings of their emotional positivity/negativity, and their focus on past/present/future across all three of the analog studies. Correlations between their attitudes about the study and about their physical well-being, as well as their focus on themselves versus others, were also repeatedly significant. Furthermore, there is evidence that automated evaluation of journal entries may do a better job of assessing participant attitudes than do daily observers: a comparative analysis between the automated assessments and subjects’ own survey responses versus the survey assessments of the subject by a nurse showed that automated analyses did a better job at predicting subjects’ survey responses.
Investigators also demonstrated the speed and flexibility of their approaches by rapidly producing multiple after-the-fact analyses of elements of participants’ attitudes and emotions throughout the study including analyses of the effects of exercise and of testosterone treatments on subjects’ emotions and attitudes about sleep and food. Further, by examining the correlates of positive or negative mentions of a word group such as food, eating, or specific food items investigators can provide suggestive evidence about how a specific individual or group thinks about that topic (that food is comforting and tied to nostalgia and social contacts or that it is a source of concern about body image and health). The flexibility to perform post hoc and deep, individualistic analyses proved to be a particularly useful capability and a strong distinguisher from survey questions.
In addition to examining journals, investigators developed techniques for examining interactive dialogue during tasks and daily activities. This form of communication is more immediate and less pre-meditated than journal entries and, therefore, is expected to be a less guarded but more transitory source of attitudes and emotions. Investigators have detected significant differences in the overall emotional valence of the different crews, of valence for differing tasks among the different crews, and of varying (and differing) drivers for those emotional sentiments among the crewmembers.
During debriefs with analog participants and operations personnel throughout this study, investigators have identified a number of recommendations for the use of the developed techniques that are particularly relevant to long-duration missions. Among these are an intervention or advisory tool for crewmembers themselves (reflecting exhibited changes in mood or attitude), a “power level” and “team comfort/routine” indicator for tracking intra-crew dynamics, and an aid for ground researchers to glean attitude and topic focus information from mission and training debriefing sessions.