Limits...
Using Graph Components Derived from an Associative Concept Dictionary to Predict fMRI Neural Activation Patterns that Represent the Meaning of Nouns.

Akama H, Miyake M, Jung J, Murphy B - PLoS ONE (2015)

Bottom Line: We measure the effectiveness of graph-based coefficients through the application of linguistic graph information for a neural activity recorded during conceptual processing in the human brain.Furthermore, correlating the voxel information with the MiF-based principal components, a new computational neurolinguistics model with a network connectivity paradigm is created.This allows two dimensions of context space to be incorporated with both semantic and neural distributional representations.

View Article: PubMed Central - PubMed

Affiliation: Graduate School of Decision Science and Technology, Tokyo Institute of Technology, Tokyo, Japan.

ABSTRACT
In this study, we introduce an original distance definition for graphs, called the Markov-inverse-F measure (MiF). This measure enables the integration of classical graph theory indices with new knowledge pertaining to structural feature extraction from semantic networks. MiF improves the conventional Jaccard and/or Simpson indices, and reconciles both the geodesic information (random walk) and co-occurrence adjustment (degree balance and distribution). We measure the effectiveness of graph-based coefficients through the application of linguistic graph information for a neural activity recorded during conceptual processing in the human brain. Specifically, the MiF distance is computed between each of the nouns used in a previous neural experiment and each of the in-between words in a subgraph derived from the Edinburgh Word Association Thesaurus of English. From the MiF-based information matrix, a machine learning model can accurately obtain a scalar parameter that specifies the degree to which each voxel in (the MRI image of) the brain is activated by each word or each principal component of the intermediate semantic features. Furthermore, correlating the voxel information with the MiF-based principal components, a new computational neurolinguistics model with a network connectivity paradigm is created. This allows two dimensions of context space to be incorporated with both semantic and neural distributional representations.

No MeSH data available.


Mean discrimination accuracies obtained from the participants of Mitchell et al.'s research [26].These results were obtained under the two MiF-EAT conditions (836 words and 60 principal components), inverse shortest path step lengths, Jaccard/Simpson indices for subsequent PCA, and the replicated results of the Google-Science paper research [26].
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4482269&req=5

pone.0125725.g005: Mean discrimination accuracies obtained from the participants of Mitchell et al.'s research [26].These results were obtained under the two MiF-EAT conditions (836 words and 60 principal components), inverse shortest path step lengths, Jaccard/Simpson indices for subsequent PCA, and the replicated results of the Google-Science paper research [26].

Mentions: We applied MiF to the EAT subgraph, and adopted the stability score to construct graph-based models from Mitchell et al.’s fMRI datasets. The precision of our decoding models was P1: 0.85, P2: 0.79, P3: 0.78, P4: 0.68, P5: 0.89, P6: 0.74, P7: 0.78, P8: 0.75, and P9: 0.76 (mean: 0.78). The corresponding results with the 60 principal components were P1: 0.87, P2: 0.75, P3: 0.76, P4: 0.66, P5: 0.89, P6: 0.72, P7: 0.76, P8: 0.70, and P9: 0.69 (mean: 0.76). The original Mitchell et al. study recorded accuracies of P1: 0.81, P2: 0.74, P3: 0.76, P4: 0.69, P5: 0.81, P6: 0.79, P7: 0.74, P8: 0.76, and P9: 0.82 (mean: 0.77). We also computed predictive models with 60 principal components extracted from the distance matrix using the inverse shortest path step lengths (mean: 0.72), Jaccard index (mean: 0.74), and Simpson index (mean: 0.75) considering the geodesic information between nodes. A non-parametric Wilcoxon signed rank test was performed between the MiF modelling result and the closest one based on the Simpson index, both with 60 principal components. The difference was found to be highly significant (p = 7.6600e-04), and MiF outperformed the other graph similarity coefficients. Figs 4 and 5 compare the participant-wise decoding accuracy and the mean discrimination accuracy of the two MiF-based EAT analysis models, inverse shortest path step lengths, Jaccard/Simpson indices for subsequent PCA, and the replicated Mitchell et al. results with the Google 5-grams Collection. Fig 6 represents an item-wise confusion matrix generated as a result of cross-validating our decoding model trained with the 60 MiF-based principal components and averaged over all nine participants. The precision in discriminating nouns is generally good, despite a slight penalty in the within-category comparisons and the cross-category ones involving the nouns of man-made objects.


Using Graph Components Derived from an Associative Concept Dictionary to Predict fMRI Neural Activation Patterns that Represent the Meaning of Nouns.

Akama H, Miyake M, Jung J, Murphy B - PLoS ONE (2015)

Mean discrimination accuracies obtained from the participants of Mitchell et al.'s research [26].These results were obtained under the two MiF-EAT conditions (836 words and 60 principal components), inverse shortest path step lengths, Jaccard/Simpson indices for subsequent PCA, and the replicated results of the Google-Science paper research [26].
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4482269&req=5

pone.0125725.g005: Mean discrimination accuracies obtained from the participants of Mitchell et al.'s research [26].These results were obtained under the two MiF-EAT conditions (836 words and 60 principal components), inverse shortest path step lengths, Jaccard/Simpson indices for subsequent PCA, and the replicated results of the Google-Science paper research [26].
Mentions: We applied MiF to the EAT subgraph, and adopted the stability score to construct graph-based models from Mitchell et al.’s fMRI datasets. The precision of our decoding models was P1: 0.85, P2: 0.79, P3: 0.78, P4: 0.68, P5: 0.89, P6: 0.74, P7: 0.78, P8: 0.75, and P9: 0.76 (mean: 0.78). The corresponding results with the 60 principal components were P1: 0.87, P2: 0.75, P3: 0.76, P4: 0.66, P5: 0.89, P6: 0.72, P7: 0.76, P8: 0.70, and P9: 0.69 (mean: 0.76). The original Mitchell et al. study recorded accuracies of P1: 0.81, P2: 0.74, P3: 0.76, P4: 0.69, P5: 0.81, P6: 0.79, P7: 0.74, P8: 0.76, and P9: 0.82 (mean: 0.77). We also computed predictive models with 60 principal components extracted from the distance matrix using the inverse shortest path step lengths (mean: 0.72), Jaccard index (mean: 0.74), and Simpson index (mean: 0.75) considering the geodesic information between nodes. A non-parametric Wilcoxon signed rank test was performed between the MiF modelling result and the closest one based on the Simpson index, both with 60 principal components. The difference was found to be highly significant (p = 7.6600e-04), and MiF outperformed the other graph similarity coefficients. Figs 4 and 5 compare the participant-wise decoding accuracy and the mean discrimination accuracy of the two MiF-based EAT analysis models, inverse shortest path step lengths, Jaccard/Simpson indices for subsequent PCA, and the replicated Mitchell et al. results with the Google 5-grams Collection. Fig 6 represents an item-wise confusion matrix generated as a result of cross-validating our decoding model trained with the 60 MiF-based principal components and averaged over all nine participants. The precision in discriminating nouns is generally good, despite a slight penalty in the within-category comparisons and the cross-category ones involving the nouns of man-made objects.

Bottom Line: We measure the effectiveness of graph-based coefficients through the application of linguistic graph information for a neural activity recorded during conceptual processing in the human brain.Furthermore, correlating the voxel information with the MiF-based principal components, a new computational neurolinguistics model with a network connectivity paradigm is created.This allows two dimensions of context space to be incorporated with both semantic and neural distributional representations.

View Article: PubMed Central - PubMed

Affiliation: Graduate School of Decision Science and Technology, Tokyo Institute of Technology, Tokyo, Japan.

ABSTRACT
In this study, we introduce an original distance definition for graphs, called the Markov-inverse-F measure (MiF). This measure enables the integration of classical graph theory indices with new knowledge pertaining to structural feature extraction from semantic networks. MiF improves the conventional Jaccard and/or Simpson indices, and reconciles both the geodesic information (random walk) and co-occurrence adjustment (degree balance and distribution). We measure the effectiveness of graph-based coefficients through the application of linguistic graph information for a neural activity recorded during conceptual processing in the human brain. Specifically, the MiF distance is computed between each of the nouns used in a previous neural experiment and each of the in-between words in a subgraph derived from the Edinburgh Word Association Thesaurus of English. From the MiF-based information matrix, a machine learning model can accurately obtain a scalar parameter that specifies the degree to which each voxel in (the MRI image of) the brain is activated by each word or each principal component of the intermediate semantic features. Furthermore, correlating the voxel information with the MiF-based principal components, a new computational neurolinguistics model with a network connectivity paradigm is created. This allows two dimensions of context space to be incorporated with both semantic and neural distributional representations.

No MeSH data available.