Named Entity Recognition and Relation Detection for Biomedical Information Extraction

Abstract

The number of scientific publications in the literature is steadily growing, containing our knowledge in the biomedical, health, and clinical sciences. Since there is currently no automatic archiving of the obtained results, much of this information remains buried in textual details not readily available for further usage or analysis. For this reason, natural language processing (NLP) and text mining methods are used for information extraction from such publications. In this paper, we review practices for Named Entity Recognition (NER) and Relation Detection (RD), allowing, e.g., to identify interactions between proteins and drugs or genes and diseases. This information can be integrated into networks to summarize large-scale details on a particular biomedical or clinical problem, which is then amenable for easy data management and further analysis. Furthermore, we survey novel deep learning methods that have recently been introduced for such tasks.

Automated Extraction of Information From Texts of Scientific Publications: Insights Into HIV Treatment Strategies

“…There were some NEs associated with HIV, such as “gag-pol protein” or “pol peptide.” Also, we found that 11 names were not associated with any proteins, they represent false-positive results of our algorithm. Therefore, the automated verification using a database or a dictionary of proteins can help filter out the named entities that represent false positive results and therefore improve the recognition accuracy obtained using CRF (Song et al, 2015 ; Perera et al, 2020 ). It also helps select the names of protein belonging to the species of interest.…”

Efficient Automated Processing of the Unstructured Documents Using Artificial Intelligence: A Systematic Literature Review and Future Directions

“…Feature extraction [10] : Various feature extraction methods used in the text classification and recognition are: GloVe, TF-IDF, and Word2Vec. The GlobalVectors (GloVe) is used to get the vector or numerical representation for words.…”

“…Lack of enough data and poor-quality data is another challenge in unstructured data processing [115]. 2) Entities related challenges: Extracting information from a highly ambiguous language such as Arabic, [43], [88], [36], [85], [40], [10], [114], [91], [88], [3], [31] especially without creating a dictionary, is challenging. The semantics and the contextual relationship among named entities for such ambiguous language are challenging for the information extraction techniques [115].…”

“…For example, "Siddharth Properties" is identified as "entity" from other text and assigned to the category "organization," and "Nilesh" is identified as "entity" and assigned to the category "person." "Named Entity Recognition and Classification" is another term used to refer to NER [10]. Business document NER is challenging because such documents may contain nonstandard phrases describing entities such as invoice_Num or invoice_No, for representing the same entity [24].…”

“…It implies that it is useful to train over more data [10], [2], [19], [11] . 4) BERT: Bidirectional Encoder Representations for Transformers (BERT) [92] , [10] , [2] , [19] , [86], is nowadays the latest word embedding approach, that is effectively used in numerous biomedical and other text mining tasks. BERT learns the text representation from both the directions to better understand the context and the relationship.…”