In this analysis, we used an word embedding model algorithm based on a two-layer neural network to conduct unsupervised learning on the conclusions of about 470,000 medical papers in the last month. and then the machine could learn the distribution form of clinical medical words. In an unoptimized system architecture, if a general quad-core processor host is used, the machine learning process takes about tens of minutes. After the machine learning is completed, various combinations of logical vocabularies can be input to quickly obtain the vocabulary response from the neural network..
In this article's example, we used five well-known clinical drugs for the treatment of cancer such as tacrolimus, enzalutamide, febuxostat, tamsulosin, and milabegron as input values, and neural networks responded to the relevant vocabulary in the conclusion of the medical paper in the last month. Response results showed that the most relevant 50 vocabulary terms are all related nouns for clinical drugs, and the vocabulary includes market competition products for treating the same cancer, functionally similar products for the treatment of similar cancers, adjuvant drugs for side effects, and products for the treatment of medical behaviors related to related cancers. Under the interpretation of senior medical editors in this field, besides the more familiar clinical treatment conclusions, there are many vocabulary combinations that have not yet understood their relevance. This result implies that in the recently published data, there are medical papers that should be thoroughly read by people in this field.
在本文的範例中，我們將五種知名的癌症治療臨床藥物 tacrolimus、enzalutamide、febuxostat、tamsulosin與mirabegron做為輸入值，使類神經網路回應最近一個月的醫學論文結論中的相關詞彙。回應結果顯示，最相關的 50個詞彙皆為臨床藥物的相關名詞，且詞彙包含了治療相同癌症的市場競爭產品、治療同類癌症的功能相似產品、緩解副作用的輔助藥物、對應相關癌症醫療行為的治療藥物等。在本領域資深的醫學編輯人員的判讀下，除了較為熟悉的臨床治療結論之外，也出現了許多尚未了解其關聯性的詞彙組合。這項結果暗示在最近發表的資料中，存在該領域人員應該深入閱讀的醫學論文。
Through this simple analysis result, we can assist medical practitioners to quickly identify relevant vocabulary that should be taken care of, and assist in follow-up data query work. Welcome to contact us, if you have any analysis needs about large-scale data.