From artificial intelligence since the beginning of the birth of the discipline, natural language processing ( NLP ) is one of the core of artificial intelligence research questions. The importance of NLP is no doubt that it is possible to implement advanced human-computer interaction in natural language communication is characterized, the machine can "read" all human knowledge recorded in text form, and provides a variety of high-level intelligence service basis and key technology.
Currently the number of Google to be the most watched in the field of NLP NLP model BERT (Bidirectional Encoder Representa-tions from Transformers), it is on the basis of Trans-former, by means of massive cross-cutting corpus and ultra-high computing power, multi-task by pre-training in more than 10 different NLP tasks to reach the highest level.
In the financial sector, the role of NLP technique include auto-discovery from the mass of the macro, industry and micro News, analysis and integration of information related to various types of decision-making (especially investment decisions), that is, first of all access to relevant information by text retrieval technology, then extracted by semantic analysis technology from unstructured text, structured information, which will be the final refine this information and make it related to possible future trends, providing valuable and timely information to decision-making and forecasting.
NLP combination of technology and machine learning techniques, is also becoming the new hot wave of financial intelligence, has been successfully applied in many scenarios, including: a smart customer service, intelligent investment research, investment adviser smart, intelligent risk control, intelligent monitoring, intelligent operations and so on.
This aspect of foreign noteworthy Applications include: Wall Street giants began to use natural language processing and machine learning techniques to help clients develop financial and retirement plans (similar financial manager role); Massachusetts Institute of Technology fusion machine learning techniques for creating new business models and redefine financial services to improve the security of investments; try an Australian financial services company uses natural language processing technology to automatically monitor and regulate the company.
Currently NLP technology, while the financial sector have been carried out in a number of attempts and exploration, but still at an early stage of development and there are some difficulties to be resolved.
BERT NLP can effectively solve the financial problems it?
Finance is highly specialized areas, many words and expressions have special meaning in the context of the financial and professional vocabulary is difficult to see get some other text. Lack of data sets in NLP is currently one of the problems facing the financial sector application, which is the high degree of professionalism in the financial sector caused.
In addition, the financial sector has its own unique understanding of events, evaluation and analysis of the results is also handled differently from other fields. Thus, natural language processing tasks in the financial sector need to be redefined mission objectives and evaluation methods, traditional and mature NLP solutions may not only simple text customized to meet the demand for information analysis and processing of the financial sector.
Then made substantial upgrade of BERT on many traditional NLP tasks, can help solve problems of financial NLP it? This is a question many people are concerned about, but for the moment may have little effect. Because BERT is designed to focus on pure and concerns expressed in natural language model itself, but the task itself or with specific areas of business closely, its contribution is smaller.
Although still need to know the exact impact of the financial sector BERT and Transformer specific tasks to experiment, but it can be expected that, due to lack of reasoning ability, aspects of input length restrictions, such as lack of interpretability congenital deficiencies, its news in the market impact assessment, event causality found that affect task-oriented context-sensitive dialogue, text summarization, intelligent recommendation and so limited.
Two ideas advanced financial NLP
And mapping knowledge , intelligent reasoning combined
The "NLP" and "knowledge map" as a financial technology in the field of "Gemini" is highly desirable idea, which coincides with the height of two key technologies in the field of financial scenarios, the two rely on each other, complement each other. The former constantly enrich the content of the latter, which provides background support, compared with the former.
不过，在“NLP+知识图谱”这两个“双子星”中，还应加上“智能推理”一环，从而形成“语义理解+知识支撑+动态推理”的“三驾马车”。因为如果模型缺乏推理能力，欠缺揭示分析结果深层原因的“可解释性”，对于金融这种需要“刨根问底”的领域容易引发“灾难后果”。
知识图谱一般认为仅存储静态的知识，静态的知识需要与动态的推理规则相结合才能推导并得出新的认识和结论，发挥所构建知识库的最大效用。虽然目前已经提出“事理图谱”概念，其是否属于动态知识仍有待商榷。引入“智能推理”可以形成事实到结论之间的推理链条，从而能够对所得结论进行必要的解释。
例如，引入类似“原材料供应紧张 → 生产成本上升 → 净利润下降 → 股价下跌”反映专业知识的规则与推理不仅可以引导模型学习的方向，缩小的搜索空间，还可以作为先验信息，进行更为合理的贝叶斯统计推断。
经验主义、理性主义缺一不可
一般认为，NLP主要有两种研究思想和方法，第一种是理性主义方法，其主要思路是通过归纳语言学规则来分析和生成语言，优点是语言表达结构和组成成分可以借助规则直接清晰地表示出来，但规则过于“刚性”会导致无法处理例外情况、鲁棒性差、规则获取和更新代价高等问题。
另一种研究方法是经验主义方法，主要是采用机器学习（特别是统计学习）从语料集中自动或半自动地获取语言学统计知识来构建模型，然后对新的文本进行推断。目前最热门的深度学习也属于经验主义方法，近年来取得了快速进展和广泛应用，在学术界和企业界备受瞩目。
对于金融领域的NLP应用来说，经验主义和理性主义这两方面不应是“离异”状态，而应该积极的“联姻”，就如哲学家培根所主张的，既反对狭隘的理性主义，也反对纯粹的经验主义。具体说来，以逻辑推理为代表的符号主义与神经网络为代表联结主义的深度融合应是未来最具发展潜力的方向之一。
Currently NLP techniques in depth understanding of natural language, there is still plenty of space exploration, such as how to accurately handle disambiguation refer to chapter within the range; how to correctly understand the analogy, metaphor and metaphor and so on. With the integration of financial NLP empiricism and rationalism enhance research methods, and further combined with the knowledge map, intelligent reasoning, expect more problems will be solved.