In the present days, research under the issue of gender and natural language processing (GenderNLP) usually target at gender-norm language that spoken by biologically males and females. However, from the standpoint of humanistic science, language is a subject to many influences like gender complexity. For this reason, the current thesis aims at exploring gendered texts from the perspective of sexual orientation in Chinese GenderNLP domain. Firstly, in order to prove that gendered texts can be well-categorized not only by biological sex, but also sexual orientation, this thesis adopts both Convolutional Neural Networks (CNNs) and Support Vector Machine (SVM) and uses both Word2Vec embeddings and linguistic feature set as input vectors to train classifiers that are able to correctly categorize texts written by homosexual males, heterosexual males, homosexual females, and heterosexual females. By simply using the threshold of 0.5 in this pilot experiment, training results show that either using CNNs or SVM, our trained classifiers are able to classify homosexual texts from heterosexual texts collect from Chinese social media PTT. Secondly, with the adoption of identical model settings as in our pilot experiment, the current thesis trains another homosexual classifier in order to automatically identify homosexual males and females’ texts. In addition, since this study expects our trained classifier does not only limits to homosexual texts records from one single source, but could also correctly classify gendered data from different textual environments, homosexual texts from two different online sources: UThome and 2Girl are also collected and used. Under the experimental limitation of time and computing resource, results of our experiment show that in such homosexual classification task, the SVM model is likely to outperform the CNNs model. Furthermore, under the linguistic analysis of homosexual texts, it is also found that gendered texts do not only differ in the use of content word, but linguistic features such as function word, punctuation, syntactic structure and statistical measurements such as lexical diversity, word count, character count, and information unpredictability also show significant statistical differences in our homosexual classification tasks.