Natural Language Processing

  • Natural language processing is an important direction in the field of computer science and artificial intelligence. ASTRI has established the core competence and technologies on NLP and applied these technologies on various application.


    Technology Highlights

    1. Chinese Natural Language Processor

    The unique cultural practice to write in proper Chinese, augmented with English, emojis, local slangs, colloquial terms, domain jargons, and intentionally or unintentionally ‘mis-written terms’ is prevalent in Hong Kong. This unprecedented ‘mix’ of linguistic expressions makes it complex to analyse the language and associated mood and emotions. ASTRI has developed natural language processing (NLP) technology embedded with a sentence parser specific for colloquial Cantonese to extract the full meaning of these written passages. It also has a built-in colloquial Cantonese vocabulary database together with a large set of synonyms, and confusion sets with words that are similar in sound or in form.

    This technology can be applied to conduct big data applications on social media analytics, including mood/sentiment analysis, topics analysis and contextual analysis. The tool can be used to assist teachers in assessing primary student’s Chinese writing and providing suggestions to correct common mistakes.

    When the NLP is connected to a voice recognition engine, it can be developed into a Chatbot application or other voice-interactive tools.



    1. Multi-lingual address recognition for smart phone

    ASTRI team has successfully developed an address recognition system that works with and implemented the regular expressions for English, French, German, Italian, Spanish, and Portuguese languages in short messages on mobile phones. By using this technology, the mobile phones are enabled to recognize addresses inside the short messages received. The accuracy and recall rate are all above 85%, verified on hundreds of thousands annotated sentences in each language. Based on this, user can easily tab the address and link to search for addresses on maps.

    ASTRI’ s Smart Investment Platform

  • The Intelligent Investment Platform, developed jointly by ASTRI and t.Axiom, is capable of analysing large-scale financial data, calculating quantitative models and processing a big chunk of historical and real-time data. It provides in-depth analysis of empirical data, ensuring efficiency and accuracy. Using artificial intelligence to analyse customer needs and attributes, it suggests practical and meaningful investment strategies to help customers with the allocation of assets and selection of investment products. The tool is used by securities companies and fund managers to quantify investments, marking a significant progress in applying financial technology in this field. Bank of China (Hong Kong) has adopted this platform –  its ‘Smart Investment’ contest was organised using this platform, in partnership with ASTRI.

    ASTRI’s Smart Investment Platform supports the users in the following ways:

    1. Intelligent stock selection and trading strategy system: To pick the stocks and automatically generate trading strategies based on big data analysis, artificial intelligence, genetic algorithm and machine learning algorithms.
    2. Graphical and intelligent trading strategy generation platform: Visual presentation of trading trends and display of back-test results to help ordinary traders to generate trading strategies in an automated manner


    Intelligent market analysis system can gauge a particular event’s impact on different assets in the market. It explains the complex cause-effect market matrix relating the customer’s performance in simple words or graphics. As the trading volumes and complexities grow, the need to deal in large chunks of data is putting an enormous amount of stress of the customers who are burdened with the task of handling an unprecedented amount of information. In addition to the complexity in information, new and evolving regulatory requirements also affect the processes and practices in the financial markets. Under such a transmuting scenario, establishing correlations between different market factors using traditional theories and models has become very difficult. Not only has the current macro-economic environment become more complex, the need for risk management and controls have become greater too. The market itself has been the result of these complex information chemical reaction presented in front of us, the information provided to us enough for him to be more in-depth, combined with mathematics, statistics and other scientific and objective analysis. Such analysis can be real-time, accurate, and even self-correcting. Major events that impact the financial standing also make the market volatile to a certain extent – depending on the time involved, the extent of the impact depends on the nature of the event itself. Therefore, the analysis of the recent major events is very important.

    ASTRI has been a pioneer in developing complex event processing system (Complex Event Process), large data processing capabilities, cloud parallel computing technology and other advanced platform and technology that have been adopted by customers for many years – from macro to micro level to carry out timely and accurate analysis, modeling, implementation of a set of processes.

    ASTRI’s tools don’t limit their stock selection capability to only the traditional stock selection strategy. Through financial indicators or technical indicators, potential actions are identified for the next time the stock value goes up or down. In the selected stock pool, by applying the corresponding trading strategies, a comprehensive entry and exit conditions and wind control mechanisms are formed. That way, we can not only target the good stocks, but also have a good trading strategy that leads to a ‘one plus one is greater than two’ outcome. Our stock trading and portfolio screening tools are best-in-class in the market – maximising the potential benefits of the transactions. We use artificial intelligence, genetic algorithms, and depth learning models to automatically search and optimise the trading strategy through self-learning and evolution, and ultimately find the most suitable approach to trading.

    Our tools look for the correlation between specific factors and trading signals through supervised learning and unsupervised learning. It will synthesize information from fundamentals, technical aspects, trading behavior, capital surface, terminal behaviour, online financial data, and third party platform information. That leads to the analysis and rigorous selection of factors that have an important impact, and the ongoing formation of an optimised pool of factors.

    In summary, ASTRI has a strong AI capability to devise technological solutions for the financial sector with a large pool of competent experts working for the institute. All of that, is meant to ensure Hong Kong’s continued leadership in the financial industry and a shining future for smart investment.

    Handheld Video Endoscope/ Laryngoscope

  • ASTRI has worked with the healthcare industry to develop a low-cost, handheld, connected Upper Gastrointestinal Endoscope. With its high definition disposable camera design, the device can stream real-time video to large displays, helping the process of diagnosis with no need for complicated sterilisation.

    Handheld video endoscope with viewer in proof-of-concept stage

    Key features:

    • Competitive cost advantage in an emerging market
    • Real time video streaming to large displays
    • High definition disposable camera design which eliminates complicated sterilisation

    Digital Pathology Analytics Platform

  • ASTRI is developing an image analytics platform for digital pathology. It is based on a scalable medical image database, distributed parallel computing engine, and a powerful application development environment for Computer-Aided Diagnosis (CAD). The platform can support efficient management of medical image data. It also facilitates reliable web-based digital pathological information exchanged in telediagnosis and agile development of CAD applications.

    Data Management System Architecture

    Key features:

    • Big data and deep learning platform
    • Agile computer aided diagnosis application development
    • Cancer screening
    • Telediagnosis

    Intelligent Cognitive Systems

  • The Intelligent Cognitive Systems Group focuses on developing machine learning and user experience technologies, and scalable solutions for smart systems, with realisation platforms ranging from embedded system to distributed cloud environments.

    In particular, the group has developed expertise in handwriting image recognition and 3D motion data analysis. This is in consideration of the fact that advanced sensor-based natural user interface is key to an effective solution the team strives to develop solutions that are user-friendly and can be operated intuitively.

    Cloud Computing

  • Built on top of multi-disciplinary research and development in network, media, and security, Cloud Computing Group (CCG) provides high performance large scale distributed computational platform with efficiency, reliability, performance and flexibility for a range of cloud computing applications from media broadcast, digital rights management, cloud storage, network virtualization, P2P webRTC (real-time communication), to intelligent demographics, Internet Finance and financial trading applications. CCG’s core computing engine builds upon innovative distributed load balancing PE (processor elements) with dynamic expansion/shrink, load shedding, and complex event processing. It has built-in scalable and distributed statistics computation algorithms aiming at multi-modal data processing, pattern recognition, and machine learning for large stream data sets. CCG team provides system, platform, solution, and consulting services so that the customers can focus on their core businesses.