Another line of my work has followed up on these subjective differences by asking whether these intrusive memories also differ in terms of objective content. Potentially, individuals in clinical populations could have been exposed the exact same events as nonclinical individuals, yet have dramatically different subjective experiences (e.g., witnessing similarly severe events, but attributing much greater emotional intensity). To address this question, I have employed computational methods to examine objective content in intrusive memory narratives, being one of the first to publish articles on analyzing autobiographical memory content using AI/ML methods and openly releasing all associated code (Yeung & Fernandes, 2022; Yeung et al., 2022).
For example, I have pioneered the use of techniques such as machine learning and natural language processing to enable large-scale analyses of what people actually report remembering, quantifying topics such as “family vacations”, “interpersonal relationship difficulties”, and “potentially traumatic events” as continuous variables in tens of thousands of written memory narratives. My work has shown that objective content still plays a significant role in predicting disorder symptoms, even after controlling for valence ratings (Yeung et al., 2022; Yeung & Fernandes, 2023); for example, content about interpersonal relationship difficulties predicted more PTSD symptoms, regardless of one’s subjective valence rating. My research argues that both subjective and objective information about intrusive memories are necessary to predict their links to mental health.
References
2023
-
Specific topics, specific symptoms: Linking the content of recurrent involuntary memories to mental health using computational text analysis
Ryan C Yeung and Myra A Fernandes
npj Mental Health Research, 2023
Researchers debate whether recurrent involuntary autobiographical memories (IAMs; memories of one’s personal past retrieved unintentionally and repetitively) are pathological or ordinary. While some argue that these memories contribute to clinical disorders, recurrent IAMs are also common in everyday life. Here, we examined how the content of recurrent IAMs might distinguish between those that are maladaptive (related to worse mental health) versus benign (unrelated to mental health). Over two years, 6187 undergraduates completed online surveys about recurrent IAMs; those who experienced recurrent IAMs within the past year were asked to describe their memories, resulting in 3624 text descriptions. Using a previously validated computational approach (structural topic modeling), we identi ed coherent topics (e.g., “Conversations” , “Experiences with family members”) in recurrent IAMs. Specific topics (e.g., “Negative past relationships”, “Abuse and trauma”) were uniquely related to symptoms of mental health disorders (e.g., depression, PTSD), above and beyond the self-reported valence of these memories. Importantly, we also found that content in recurrent IAMs was distinct across symptom types (e.g., “Communication and miscommunication” was related to social anxiety, but not symptoms of other disorders), suggesting that while negative recurrent IAMs are transdiagnostic, their content remains unique across different types of mental health concerns. Our work shows that topics in recurrent IAMs—and their links to mental health—are identifiable, distinguishable, and quantifiable.
2022
-
Machine learning to detect invalid text responses: Validation and comparison to existing detection methods
Ryan C Yeung and Myra A Fernandes
Behavior Research Methods, 2022
A crucial step in analysing text data is the detection and removal of invalid texts (e.g., texts with meaningless or irrelevant content). To date, research topics that rely heavily on analysis of text data, such as autobiographical memory, have lacked methods of detecting invalid texts that are both effective and practical. Although researchers have suggested many data quality indicators that might identify invalid responses (e.g., response time, character/word count), few of these methods have been empirically validated with text responses. In the current study, we propose and implement a supervised machine learning approach that can mimic the accuracy of human coding, but without the need to hand-code entire text datasets. Our approach (a) trains, validates, and tests on a subset of texts manually labelled as valid or invalid, (b) calculates performance metrics to help select the best model, and (c) predicts whether unlabelled texts are valid or invalid based on the text alone. Model validation and evaluation using autobiographical memory texts indicated that machine learning accurately detected invalid texts with performance near human coding, significantly outperforming existing data quality indicators. Our openly available code and instructions enable new methods of improving data quality for researchers using text as data.
-
Understanding autobiographical memory content using computational text analysis
Ryan C Yeung, Marek Stastna, and Myra A Fernandes
Memory, 2022
Although research on autobiographical memory (AM) continues to grow, there remain few methods to analyze AM content. Past approaches are typically manual, and prohibitively time- and labour-intensive. These methodological limitations are concerning because content may provide insights into the nature and functions of AM. In particular, analyzing content in recurrent involuntary autobiographical memories (IAMs; those that spring to mind unintentionally and repetitively) could resolve controversies about whether these memories typically involve mundane or distressing events. Here, we present computational methods that can analyze content in thousands of participants’ AMs, without needing to hand-code each memory. A sample of 6,187 undergraduates completed surveys about recurrent IAMs, resulting in 3,624 text descriptions. Using frequency analyses, we identified common (e.g., “time” , “friend”) and distinctive words in recurrent IAMs (e.g., “argument” as distinctive to negative recurrent IAMs). Using structural topic modelling, we identified coherent topics (e.g., “Negative past relationships” , “Conversations” , “Experiences with family members”) within recurrent IAMs and found that topic use significantly differed depending on the valence of these memories. Computational methods allowed us to analyze large quantities of AM content with enhanced granularity and reproducibility. We present the means to enable future research on AM content at an unprecedented scope and scale.