Editor's introduction: “no knowledge mapping, no information”, knowledge mapping has always been a subject of great discussion in the field of ai, with major companies building knowledge mapping in their respective fields, and many unicorn companies, which rely on knowledge mapping. So what is the knowledge map? Why is it so popular? How is it different from the traditional way in which data are stored in databases? Let's see。
I. What is the knowledge map
As literally understood, the knowledge map can be broken down into “knowledge” and “specify” to understand。
Knowledge is the sum of the knowledge and experience acquired in the practice of transforming the world。
Graphics are a way to store links。
Knowledge mapping is actually a way to store and express knowledge。
But why is a way of storing a knowledge being singled out? Even being singled out by google as a technology project
It is first necessary to understand the role of knowledge in awareness。
Indeed, prior to google, the knowledge map was a precursor: semantic networks were introduced as early as the 1960s, as a way of expressing knowledge, mainly in the field of natural language understanding (making the machine understand language)。
Important nodes of knowledge mapping evolution
Just as one understands what is expressed in a statement, one uses knowledge, such as “wei has a newly released apple in his hand”, a phrase that needs to be understood correctly, and at least knows that apples can be a mobile phone sign, in addition to food。
That is to say, the origin of the knowledge mapping is to give machines a better understanding of the language and, by building a knowledge base, to try to give machines more “knowledge points”, more connections and reasoning and better understanding of information。
Knowledge mapping as a role of memory in artificial intelligence
The role of knowledge is not limited to text, but also to pictures。
It's funny to read that it's funny, it's funny, it's funny, it's funny, it's funny to think about the image of actors in social media, what they look like, how they act, how they think about their relationship。
Ii. Why maps
Why has it evolved to this stage to store knowledge through mapping rather than traditional databases
I think there are two things:
The first is because the storage structure of the spectra is simple enough, and the spectra is stored through a “triple-chamber group”, i. E., the head entity, the relationship and the composition of the tail entity
For example, the apple phone is a product under the umbrella of apple, and the three-tier group, which is an abstract expression of knowledge, is:
Almost all knowledge that can be expressed through symbols can be stored using this structure。
The second is the spread of interconnective thinking, which, in an era of emphasis on the interconnection of everything, can be stored in such a way as to focus on links that can link knowledge in different areas and thus exploit the hidden value of relationships。
Iii. What is the use of knowledge mapping
There are two main directions for current knowledge mapping applications:
1. Complementary language understanding
The role of knowledge mapping in supporting language understanding is as follows:
(c) substantive differences: a precise determination of the multiple meanings mentioned in the text, as in the apple case mentioned above。
Indication: an explanation of the pronoun in the text, such as him and it。
The representative applications are as follows:
(1) search
The traditional search only provides a search of the web page (the red box section), and the graphics provide a description of the thing itself, making the results more intuitive and corresponding to the semantic of the query。
Spectrum search effects presentation
(2) questions and answers
The question-and-answer system in the vertical field involves many issues of expertise, for example in the insurance sector:
For example, when clients ask, "is xx insurance safe for polio?"
Knowledge mapping can provide precise answers by combining knowledge in the field of insurance with medical knowledge。
Simplification of knowledge reasoning
2. Supplementary big data analysis
The mapping, which combines knowledge in a variety of fields and creates an area-specific knowledge map, is currently used mainly in the industrial landings in terms of large data analysis, as follows:
(1) recommendations
The knowledge map contains a wealth of relevance and can provide a partial source of information for the referral system; for example, commonly recommended by film and music。
The benefits of joining the mapping recommendations are primarily interpretable and accurate delivery based on a well-defined recommended path。
Ding likes emily. Emily's band is back in spring, so the same band can be recommended。
(2) wind control
The structure of the map is very well combined with the sna (social network analysis) theory, and the risk of group fraud can be exploited very well; for example, money-laundering can combine multiple transactions, business information, etc. To see whether there is a pool of funds。
The advantage of the figure is that it can be calculated over multiple degrees, that it can exploit a deeper risk relationship and that it is more comprehensive than traditional surveillance techniques to map the risk of groups operating。
Iii. No map, no ai
At this stage, it would be nice if you understood the pr version, and in fact many ai applications are not used in knowledge mapping, which itself has considerable limitations。
1. Industrial landing perspective
At least at this stage, there are many options for the replacement of knowledge mapping in many projects, such as a wind-controlled scene, which can capture many of the cases through traditional data analysis; and a question-and-answer landscape, with a simple knowledge base, which is more efficient and less input than a direct knowledge mapping。
The reason is also that this technology requires a very large investment of resources, a large number of experts with substantial operational knowledge, algorithms, etc。
Almost 70 per cent of the actual industrial landing projects are devoted to the acquisition, cleansing and structuring of mapping data, while less than 30 per cent of the time is devoted to the construction and application of knowledge frameworks。
2. Technology perspective
As mentioned above, the three-dimensional form of the knowledge map represents almost all the symbolic knowledge, that is, the explicit knowledge that can be well expressed, but there is a lot of knowledge in reality that is hidden, such as a skill that playing the piano light knows the music but does not have a good finger。
While knowledge mapping requires a very high degree of data structure, most business data in reality are unstructured and how to move from unstructured data to structured, which is an unsolved industry challenge and a bottleneck for nlps。
Concluding remarks
For each new technology, we should see the space for its future development, as well as its present limitations。
Understanding, reflection and reflection on technology are essential if technology is to be truly put on the ground。
