Center for Learning and Experimental Psychopathology.
School of Psychology, Victoria University of Wellington.
Psychol Assess. 2018 Feb;30(2):259-273. doi: 10.1037/pas0000472. Epub 2017 Apr 3.
Reduced specificity of autobiographical memories is a hallmark of depressive cognition. Autobiographical memory (AM) specificity is typically measured by the Autobiographical Memory Test (AMT), in which respondents are asked to describe personal memories in response to emotional cue words. Due to this free descriptive responding format, the AMT relies on experts' hand scoring for subsequent statistical analyses. This manual coding potentially impedes research activities in big data analytics such as large epidemiological studies. Here, we propose computerized algorithms to automatically score AM specificity for the Dutch (adult participants) and English (youth participants) versions of the AMT by using natural language processing and machine learning techniques. The algorithms showed reliable performances in discriminating specific and nonspecific (e.g., overgeneralized) autobiographical memories in independent testing data sets (area under the receiver operating characteristic curve > .90). Furthermore, outcome values of the algorithms (i.e., decision values of support vector machines) showed a gradient across similar (e.g., specific and extended memories) and different (e.g., specific memory and semantic associates) categories of AMT responses, suggesting that, for both adults and youth, the algorithms well capture the extent to which a memory has features of specific memories. (PsycINFO Database Record
自传体记忆特异性降低是抑郁认知的一个标志。自传体记忆 (AM) 特异性通常通过自传体记忆测试 (AMT) 来衡量，在该测试中，要求受访者用情绪提示词来描述个人记忆。由于这种自由描述的反应形式，AMT 依赖于专家的手工评分进行后续的统计分析。这种手动编码可能会阻碍大数据分析（如大型流行病学研究）等研究活动。在这里，我们提出了计算机算法，通过自然语言处理和机器学习技术，自动对 AMT 的荷兰语（成年参与者）和英语（青年参与者）版本进行 AM 特异性评分。这些算法在独立测试数据集（接收者操作特征曲线下的面积>0.90）中可靠地区分了特定和非特定（例如过度泛化）的自传体记忆。此外，算法的结果值（即支持向量机的决策值）在类似（例如特定和扩展记忆）和不同（例如特定记忆和语义联想）的 AMT 反应类别中呈现出梯度，这表明对于成年人和年轻人来说，算法很好地捕捉到了记忆具有特定记忆特征的程度。
