Deep learning has been a mainstream solution for recognizing printed and isolated handwritten characters in Optical Characters Recognition (OCR). However, it is still a challenge to hybrid recognition of adjoining strings in printed and handwritten format, especially in the case that characters are touching and the data is imbalanced. In this paper, we propose a hybrid recognition scheme, termed TriView-ParNet, for adjoining printed-and-handwritten strings. First of all, we introduce a Parallel Network, which consists of Two-stream feature Extraction and Fusion Module (TEFM) and Context Extraction and Transcription Module (CETM). The TEFM is proposed to address the issue where characters are touched in printed and handwritten format. It can fuse the content and positional features extracted by two feature extraction networks to enrich the original feature representation. For another, the CETM is used to further extract the contextual information of the sequence. By using the contextual prompts of sequence, the recognition ability of long strings can be enhanced by CETM. Secondly, we propose a Three-view Co-training Module, in view of the poor performance of direct training based on a small amount of labeled data. Using the idea of semi-supervised learning, a classifier is trained from three different views, print, handwriting, and hybrid. Finally, we compare our method with state-of-the-art methods on the public dataset NIST SD19 and the newly collected dataset CPHS2020. The experimental results demonstrate that our method gets a higher accuracy of strings recognition. As a result, our TriView-ParNet extracts positional and contextual information to enhance the performance of recognition, which also provides a semi-supervised learning solution.