- Abstract: Academic homepages are important channels for learning researchers' profiles. Knowing the person names in academic homepages is essential to the extraction of other entities such as contacts, publications, and biography. Traditional NER models are trained on newswire corpora such as CoNLL-2003, which contain well-formed names in consistent and complete syntax. However, academic homepages often contain text with incomplete syntax and names of various forms. Few studies have addressed person name recognition in this context. To fill this gap, we propose a new name annotation scheme which provides detailed name form information including first, middle, or last name, and a full name word or a name initial. To take full advantage of this annotation scheme, we propose a Co-guided Neural Network (CogNN) model that can accurately recognise person names in academic homepages using the fine-grained annotations. CogNN uses co-attention mechanism to co-guide two jointly trained neural networks. Experimental results on real datasets show that CogNN significantly outperforms state-of-the-art NER models in extracting person names from academic homepages, while achieving comparable performance on a traditional NER benchmark dataset.