Diese Datenbank enthält über 40.000 Dokumente zu Themen aus den Bereichen Formalerschließung – Inhaltserschließung – Information Retrieval.
© 2015 W. Gödert, TH Köln, Institut für Informationswissenschaft / Powered by litecat, BIS Oldenburg (Stand: 21. Januar 2019)
1Lu, K. ; Mao, J. ; Li, G.: Toward effective automated weighted subject indexing : a comparison of different approaches in different environments.
In: Journal of the Association for Information Science and Technology. 69(2018) no.1, S.121-133.
Abstract: Subject indexing plays an important role in supporting subject access to information resources. Current subject indexing systems do not make adequate distinctions on the importance of assigned subject descriptors. Assigning numeric weights to subject descriptors to distinguish their importance to the documents can strengthen the role of subject metadata. Automated methods are more cost-effective. This study compares different automated weighting methods in different environments. Two evaluation methods were used to assess the performance. Experiments on three datasets in the biomedical domain suggest the performance of different weighting methods depends on whether it is an abstract or full text environment. Mutual information with bag-of-words representation shows the best average performance in the full text environment, while cosine with bag-of-words representation is the best in an abstract environment. The cosine measure has relatively consistent and robust performance. A direct weighting method, IDF (Inverse Document Frequency), can produce quick and reasonable estimates of the weights. Bag-of-words representation generally outperforms the concept-based representation. Further improvement in performance can be obtained by using the learning-to-rank method to integrate different weighting methods. This study follows up Lu and Mao (Journal of the Association for Information Science and Technology, 66, 1776-1784, 2015), in which an automated weighted subject indexing method was proposed and validated. The findings from this study contribute to more effective weighted subject indexing.
Inhalt: Vgl.: http://onlinelibrary.wiley.com/doi/10.1002/asi.23912/full.
Anmerkung: Vgl. das Erratum in JASIST 69(2018) no.7, S.956.
Themenfeld: Automatisches Indexieren ; Indexierungsstudien
2Mao, J. ; Cui, H.: Identifying bacterial biotope entities using sequence labeling : performance and feature analysis.
In: Journal of the Association for Information Science and Technology. 69(2018) no.9, S.1134-1147.
Abstract: Habitat information is important to biodiversity conservation and research. Extracting bacterial biotope entities from scientific publications is important to large scale study of the relationships between bacteria and their living environments. To facilitate the further development of robust habitat text mining systems for biodiversity, following the BioNLP task framework, three sequence labeling techniques, CRFs (Conditional Random Fields), MEMM (Maximum Entropy Markov Model) and SVMhmm (Support Vector Machine) and one classifier, SVMmulticlass, are compared on their performance in identifying three types of bacterial biotope entities: bacteria, habitats and geographical locations. The effectiveness of a variety of basic word formation features, syntactic features, and semantic features are exploited and compared for the three sequence labeling methods. Experiments on two publicly available BioNLP collections show that, in addition to a WordNet feature, word embedding featured clusters (although not trained with the task-specific corpus) consistently improve the performance for all methods on all entity types in both collections. Other features produce various results. Our results also show that when trained on limited corpora, Brown clusters resulted in better performance than word embedding clusters did. Further analysis suggests that the entity recognition performance can be greatly boosted through improving the accuracy of entity boundary identification.
Inhalt: Vgl.: https://onlinelibrary.wiley.com/doi/10.1002/asi.24032.
3Lu, K. ; Mao, J.: ¬An automatic approach to weighted subject indexing : an empirical study in the biomedical domain.
In: Journal of the Association for Information Science and Technology. 66(2015) no.9, S.1776-1784.
Abstract: Subject indexing is an intellectually intensive process that has many inherent uncertainties. Existing manual subject indexing systems generally produce binary outcomes for whether or not to assign an indexing term. This does not sufficiently reflect the extent to which the indexing terms are associated with the documents. On the other hand, the idea of probabilistic or weighted indexing was proposed a long time ago and has seen success in capturing uncertainties in the automatic indexing process. One hurdle to overcome in implementing weighted indexing in manual subject indexing systems is the practical burden that could be added to the already intensive indexing process. This study proposes a method to infer automatically the associations between subject terms and documents through text mining. By uncovering the connections between MeSH descriptors and document text, we are able to derive the weights of MeSH descriptors manually assigned to documents. Our initial results suggest that the inference method is feasible and promising. The study has practical implications for improving subject indexing practice and providing better support for information retrieval.
Inhalt: Vgl.: http://onlinelibrary.wiley.com/doi/10.1002/asi.23290/abstract.
Themenfeld: Indexierungsstudien ; Automatisches Indexieren
4Mao, J. ; Xu, W. ; Yang, Y. ; Wang, J. ; Yuille, A.L.: Explain images with multimodal recurrent neural networks.
Abstract: In this paper, we present a multimodal Recurrent Neural Network (m-RNN) model for generating novel sentence descriptions to explain the content of images. It directly models the probability distribution of generating a word given previous words and the image. Image descriptions are generated by sampling from this distribution. The model consists of two sub-networks: a deep recurrent neural network for sentences and a deep convolutional network for images. These two sub-networks interact with each other in a multimodal layer to form the whole m-RNN model. The effectiveness of our model is validated on three benchmark datasets (IAPR TC-12 , Flickr 8K , and Flickr 30K ). Our model outperforms the state-of-the-art generative method. In addition, the m-RNN model can be applied to retrieval tasks for retrieving images or sentences, and achieves significant performance improvement over the state-of-the-art methods which directly optimize the ranking objective function for retrieval.
Themenfeld: Automatisches Indexieren
Behandelte Form: Bilder