DASentimental: Detecting Depression, Anxiety, and Stress in Texts via Emotional Recall, Cognitive Networks, and Machine Learning

Most current affect scales and sentiment analysis on written text focus on quantifying valence/sentiment, the primary dimension of emotion. Distinguishing broader, more complex negative emotions of similar valence is key to evaluating mental health. We propose a semi-supervised machine learning model, DASentimental, to extract depression, anxiety, and stress from written text. We trained DASentimental to identify how N = 200 sequences of recalled emotional words correlate with recallers’ depression, anxiety, and stress from the Depression Anxiety Stress Scale (DASS-21). Using cognitive network science, we modeled every recall list as a bag-of-words (BOW) vector and as a walk over a network representation of semantic memory—in this case, free associations. This weights BOW entries according to their centrality (degree) in semantic memory and informs recalls using semantic network distances, thus embedding recalls in a cognitive representation. This embedding translated into state-of-the-art, cross-validated predictions for depression (R = 0.7), anxiety (R = 0.44), and stress (R = 0.52), equivalent to previous results employing additional human data. Powered by a multilayer perceptron neural network, DASentimental opens the door to probing the semantic organizations of emotional distress. We found that semantic distances between recalls (i.e., walk coverage), was key for estimating depression levels but redundant for anxiety and stress levels. Semantic distances from “fear” boosted anxiety predictions but were redundant when the “sad–happy” dyad was considered. We applied DASentimental to a clinical dataset of 142 suicide notes and found that the predicted depression and anxiety levels (high/low) corresponded to differences in valence and arousal as expected from a circumplex model of affect. We discuss key directions for future research enabled by artificial intelligence detecting stress, anxiety, and depression in texts.

The intrinsic spatiotemporal organization of the human brain - A multi-dimensional functional network atlas

The human brain is a complex, integrative and segregative network that exhibits dynamic fluctuations in activity across space and time. A canonical set of large-scale networks has been historically identified from resting-state fMRI (rs-fMRI), including the default mode, visual, somatomotor, salience, attention, and executive control. However, the methods used in identification of these networks have relied on assumptions that may inadvertently constrain their properties and consequently our understanding of the human connectome. Here we define a brain "network" as a functional component that jointly describes its spatial distribution and temporal dynamics, where neither domain suffers from unrealistic constraints. Using our recently developed BrainSync algorithm and the Nadam-Accelerated SCAlable and Robust (NASCAR) tensor decomposition, we identified twenty-three brain networks using rs-fMRI data from a large group of healthy subjects acquired by the Human Connectome Project. These networks are spatially overlapped, temporally correlated, and highly reproducible across two independent groups and sessions. We show that these networks can be clustered into six distinct functional categories and naturally form a representative functional network atlas for a healthy population. Using this atlas, we demonstrate that individuals with attention-deficit/hyperactivity disorder display disproportionate brain activity increases, relative to neurotypical subjects, in visual, auditory, and somatomotor networks concurrent with decreases in the default mode and higher-order cognitive networks. Thus, this work not only yields a highly reproducible set of spatiotemporally overlapped functional brain networks, but also provides convergent evidence that individual differences in these networks can be used to explain individual differences in neurocognitive functioning.

What Do Cognitive Networks Do? Simulations of Spoken Word Recognition Using the Cognitive Network Science Approach

Cognitive network science is an emerging approach that uses the mathematical tools of network science to map the relationships among representations stored in memory to examine how that structure might influence processing. In the present study, we used computer simulations to compare the ability of a well-known model of spoken word recognition, TRACE, to the ability of a cognitive network model with a spreading activation-like process to account for the findings from several previously published behavioral studies of language processing. In all four simulations, the TRACE model failed to retrieve a sufficient number of words to assess if it could replicate the behavioral findings. The cognitive network model successfully replicated the behavioral findings in Simulations 1 and 2. However, in Simulation 3a, the cognitive network did not replicate the behavioral findings, perhaps because an additional mechanism was not implemented in the model. However, in Simulation 3b, when the decay parameter in spreadr was manipulated to model this mechanism the cognitive network model successfully replicated the behavioral findings. The results suggest that models of cognition need to take into account the multi-scale structure that exists among representations in memory, and how that structure can influence processing.

A Survey on Traffic Prediction Techniques Using Artificial Intelligence for Communication Networks

Much research effort has been conducted to introduce intelligence into communication networks in order to enhance network performance. Communication networks, both wired and wireless, are ever-expanding as more devices are increasingly connected to the Internet. This survey introduces machine learning and the motivations behind it for creating cognitive networks. We then discuss machine learning and statistical techniques to predict future traffic and classify each into short-term or long-term applications. Furthermore, techniques are sub-categorized into their usability in Local or Wide Area Networks. This paper aims to consolidate and present an overview of existing techniques to stimulate further applications in real-world networks.

Overt speech critically changes lateralization index and did not allow determination of hemispheric dominance for language: an fMRI study

Background Pre-surgical mapping of language using functional MRI aimed principally to determine the dominant hemisphere. This mapping is currently performed using covert linguistic task in way to avoid motion artefacts potentially biasing the results. However, overt task is closer to natural speaking, allows a control on the performance of the task, and may be easier to perform for stressed patients and children. However, overt task, by activating phonological areas on both hemispheres and areas involved in pitch prosody control in the non-dominant hemisphere, is expected to modify the determination of the dominant hemisphere by the calculation of the lateralization index (LI). Objective Here, we analyzed the modifications in the LI and the interactions between cognitive networks during covert and overt speech task. Methods Thirty-three volunteers participated in this study, all but four were right-handed. They performed three functional sessions consisting of (1) covert and (2) overt generation of a short sentence semantically linked with an audibly presented word, from which we estimated the “Covert” and “Overt” contrasts, and a (3) resting-state session. The resting-state session was submitted to spatial independent component analysis to identify language network at rest (LANG), cingulo-opercular network (CO), and ventral attention network (VAN). The LI was calculated using the bootstrapping method. Results The LI of the LANG was the most left-lateralized (0.66 ± 0.38). The LI shifted from a moderate leftward lateralization for the Covert contrast (0.32 ± 0.38) to a right lateralization for the Overt contrast (− 0.13 ± 0.30). The LI significantly differed from each other. This rightward shift was due to the recruitment of right hemispheric temporal areas together with the nodes of the CO. Conclusion Analyzing the overt speech by fMRI allowed improvement in the physiological knowledge regarding the coordinated activity of the intrinsic connectivity networks. However, the rightward shift of the LI in this condition did not provide the basic information on the hemispheric language dominance. Overt linguistic task cannot be recommended for clinical purpose when determining hemispheric dominance for language.

Modeling of the processes of extracting rules from Neural Network components with Petri nets

The paper deals with the use of extended Petri nets in modeling the processes of extracting rules from neural network components. The mathematical model for extracting rules from neural network components based on a modified timed Petri net is constructed, followed by an analysis of its dynamic behavior based on a timed reachability graph, which is a set of all its states that can be reached when a finite number of transitions are fired. The proposed model allows us to move from the initial detailed structure to its simplified description, which preserves the possibility of obtaining information about the structure and dynamic behavior of the neural network system. The proposed approach can be used in the synthesis of cognitive systems with a neural network organization to provide computational support for the functions of forming, learning, and correcting cognitive networks that display neural network models.

Synthetic criticality in cellular brains

Cognitive networks have evolved to cope with uncertain environments in order to make reliable decisions. Such decision making circuits need to respond to the external world in efficient and flexible ways, and one potentially general mechanism of achieving this is grounded in critical states. Mounting evidence has shown that brains operate close to such critical boundaries consistent with self-organized criticality (SOC). Is this also taking place in small-scale living systems, such as cells? Here, we explore a recent model of engineered gene networks that have been shown to exploit the feedback between order and control parameters (as defined by expression levels of two coupled genes) to achieve an SOC state. We suggest that such SOC motif could be exploited to generate adaptive behavioral patterns and might help design fast responses in synthetic cellular and multicellular organisms.

Cognitive networks identify dimensions of distress in suicide notes: Anxiety, emotional profiles, and "words not said"

Suicide remains a serious public-health concern that is difficult to accurately predict in real-world settings. To identify potential predictors of suicide, we examined the emotional content of suicide notes using methods from cognitive network science. Specifically, we compared the co-occurrence networks of suicide notes with those constructed out of emotion words written by individuals scoring low or high on measures of depression, anxiety, and stress. Our objective was to identify which networks were most similar to the suicide notes network, in particular with regard to the connectivity between words and their emotional contents. We also investigated what types of words remained in the high/low emotion networks after controlling for the words present in the suicide notes, which we conceptualize as the “words not said” in the suicide notes. We found that patterns of connectivity among emotion words in suicide notes were most similar to those in texts written by low-anxiety individuals. However, upon analyzing the “words not said” in suicide notes, we observed that the remaining collection of emotions in suicide notes was most similar to those expressed by high-anxiety individuals. We discuss how these findings relate with existing clinical psychological literature as well as their potential implications for predicting suicidal behavior.

Functional connectivity alteration of the deprived auditory regions with cognitive networks in deaf and inattentive adolescents

Adolescents with early profound deafness may present with distractibility and inattentiveness. The brain mechanisms underlying these attention impairments remain unclear. We performed resting-state functional magnetic resonance imaging to investigate the functional connectivity of the superior temporal and transverse temporal gyri in 25 inattentive adolescents with bilateral prelingual profound deafness, and compared the results with those of 27 age-matched normal controls. Pearson and Spearman’s rho correlation analyses were used to investigate the correlations of altered functional connectivity with the attention scores on the Montreal Cognitive Assessment, years of deafness, sign language, and hearing aid usage. Compared with normal controls, prelingual profound deafness demonstrated mainly decreased resting-state functional connectivity between the deprived auditory regions and several other brain functional networks, including the attention control, language comprehension, default-mode, and sensorimotor networks. Moreover, we also found enhanced resting-state functional connectivity between the deprived auditory cortex and salience network. These results indicate a negative impact of early hearing loss on the attentional and other high cognitive networks, and the use of sign language and hearing aids normalized the participants’ connectivity between the primary auditory cortex and attention networks, which is crucial for the early intervention and clinical care of deaf adolescents.