Securities Knowledge Model
Overview
The Securities Knowledge Model (display name: Securities EN v#) categorizes English texts according to the security types they are about and extracts information about securities, related entities, values and relevant trends.
Categorization
This is the category tree:
1000 Securities
1100 Equity
1110 Private
1120 Public
1200 Debt
1210 Corporate
1220 Government
1230 Bank
1300 Hybrids
1310 Convertible bonds
1320 Reference shares
1400 Derivatives
1410 Options
1420 Swaps
1430 Futures and forwards
Securities is predicted when the text is about securities in the absence of more specific topics.
Equity branch includes:
- Private: not publicly traded equities.
- Public: stock in publicly traded companies (PLCs).
Debt branch is for debt securities or bonds. It includes:
- Corporate: bonds issued by companies.
- Government: bonds issued by national, regional, or municipal authorities.
- Banks: bonds issued by banks.
Hybrids branch is for securities that share features of both equity and debt securities. It includes:
- Reference shares: shares of a company's stock with dividends that are paid out to shareholders before common stock dividends are issued.
- Convertible bonds: fixed-income corporate debt security that can be converted into a predetermined quantity of equity.
Derivatives branch is about securities based on other underlying financial products such as other securities, commodities, currencies, etc. It includes:
- Options: derivative contracts that give the buyer the right, but not the obligation, to buy or sell the underlying asset at a specified price and by a certain date.
- Swaps: derivative contracts through which two parties exchange the cash flows or liabilities from two different financial instruments.
- Futures and forwards: derivative contracts that involve two parties who agree to buy or sell the underlying asset at a specified price by a certain date in the future.
Extraction groups and classes
The extraction part goal is to extract securities, their issuing or related entity, value, and relevant trends on the securities market, including uptrend, downtrend, earning, overbuying, and overselling.
As each security type has different features, a variety of trends and values are mentioned when dealing with securities.
Given the heterogeneity of the topic and the resources available for the project, the extraction groups include values and trends shared by most security products. This means, for instance, that private equity values are not represented in the extraction corpus, since for its own nature, information about private equity is not publicly disclosed.
Similarly, specific values related to the derivatives market (such as open interest, volume, puts and call for options, specific tickers for futures contracts…) were to be disregarded.
The extraction part includes the following groups and related classes:
SECURITIES and SECURITIES_EARNING
The SECURITIES and SECURITIES_EARNING groups have the goal to extract data about securities, issuing entities, values and security earning values (be it dividend, interest rate yield or return, according to the security type).
Their classes are:
-
SECURITIES:
- security
- entity
- value
-
SECURITY_EARNING:
- security_earns
- entity_earns
- dividend_interest_yield
- dividend_interest_yield_value
SECURITY_UP, SECURITY_DOWN, SECURITY_OVERSOLD and SECURITY_OVERBOUGHT
The SECURITY_UP, SECURITY_DOWN, SECURITY_OVERSOLD, SECURITY_OVERBOUGHT groups extract data about securities showing a particular trend together with their issuing entity if available.
Moreover, the SECURITY_UP and SECURITY_DOWN groups also include specific classes for golden cross and death cross trends, as well as a class to detect inferences, when a security/entity is mentioned in a sentence and its trend in the following one.
-
SECURITY_UP:
- security_up
- entity_up
- up_by
- security_golden_cross
- entity_golden_cross
- inference_security_by
-
SECURITY_DOWN:
- security_down
- entity_down
- down_by
- security_death_cross
- entity_death_cross
- inference_security_by
-
SECURITY_OVERSOLD:
- security_oversold
- entity_oversold
-
SECURITY_OVERBOUGHT
- security_overbought
- entity_overbought
Output structure
The model output has the same structure as any other model and is affected by the functional properties of the workflow block.
The peculiar parts of the output are the result of categorization, i.e. the categories array, and the result of information extraction, i.e. the extractions array.
Example
Considering the text:
SenseTime to Price Shares at HK$3.85 in Hong Kong IPO - Sources
HONG KONG (Reuters) -Chinese artificial intelligence start-up SenseTime Group plans to price its shares at HK$3.85 ($0.4936) each to raise $740 million in its revised Hong Kong initial public offering (IPO), according to two sources with direct knowledge of the matter.
The sources could not be named as the information has not been made public.
SenseTime declined to comment to Reuters.
The company sold 1.5 billion shares in the deal, its second attempt to list in Hong Kong in a matter of weeks. It had flagged a HK$3.85 to $HK3.99 per share price range when the deal was relaunched on Monday. Its initial attempt was shelved after SenseTime was placed on a U.S. investment blacklist as the institutional bookbuild was being finalised. The decision by the U.S. Treasury meant U.S. investors could not buy into the offering which was withdrawn while SenseTime revised its prospectus. SenseTime increased the value of the cornerstone investors from $450 million to about $511 million in a bid to have the IPO carried out successfully. Among its nine cornerstone investors are China's state-owned Mixed Ownership Reform Fund, which committed the largest portion of $200 million, Shanghai government-backed Xuhui Capital and Shanghai Guosheng Group, state-owned carmaker SAIC Motor Corp Ltd and Hong Kong government fund HKSTP Venture Fund. Hong Kong-listed ophthalmic services provider C-MER is the only non-state cornerstone investor. In its revised prospectus, SenseTime said the investment ban would have no impact on its business operations. SenseTime shares are due to start trading on the Hong Kong Stock Exchange on December 30.
the categorization output is:
"categories": [
{
"frequency": 0,
"hierarchy": [
"Securities",
"Equity",
"Public"
],
"id": "1120",
"label": "Public",
"namespace": "securities",
"positions": [...],
"score": 50,
"winner": true
}
]
In the extraction output, the template key corresponds to the concept of group and template fields correspond to classes:
"extractions": [
{
"fields": [
{
"name": "entity",
"positions": [
{
"end": 9,
"start": 0
},
{
"end": 1619,
"start": 1610
},
{
"end": 419,
"start": 410
},
{
"end": 959,
"start": 950
}
],
"value": "SenseTime Group"
},
{
"name": "security",
"positions": [
{
"end": 25,
"start": 19
},
{
"end": 1626,
"start": 1620
}
],
"value": "share"
},
{
"name": "value",
"positions": [
{
"end": 36,
"start": 31
}
],
"value": "3.85 dollar"
}
],
"namespace": "securities",
"template": "SECURITIES"
},
{
"fields": [
{
"name": "entity_up",
"positions": [
{
"end": 9,
"start": 0
}
],
"value": "SenseTime Group"
},
{
"name": "inference_security_by",
"positions": [
{
"end": 167,
"start": 161
}
],
"value": "share"
},
{
"name": "security_up",
"positions": [
{
"end": 25,
"start": 19
}
],
"value": "share"
},
{
"name": "up_by",
"positions": [
{
"end": 178,
"start": 173
}
],
"value": "3.85 dollar"
}
],
"namespace": "securities",
"template": "SECURITY_UP"
},
{
"fields": [
{
"name": "entity_up",
"positions": [
{
"end": 9,
"start": 0
}
],
"value": "SenseTime Group"
},
{
"name": "inference_security_by",
"positions": [
{
"end": 167,
"start": 161
}
],
"value": "share"
},
{
"name": "security_up",
"positions": [
{
"end": 25,
"start": 19
}
],
"value": "share"
},
{
"name": "up_by",
"positions": [
{
"end": 187,
"start": 180
}
],
"value": "0.4936 dollar"
}
],
"namespace": "securities",
"template": "SECURITY_UP"
},
{
"fields": [
{
"name": "entity_up",
"positions": [
{
"end": 9,
"start": 0
}
],
"value": "SenseTime Group"
},
{
"name": "inference_security_by",
"positions": [
{
"end": 167,
"start": 161
}
],
"value": "share"
},
{
"name": "security_up",
"positions": [
{
"end": 25,
"start": 19
}
],
"value": "share"
},
{
"name": "up_by",
"positions": [
{
"end": 215,
"start": 203
}
],
"value": "740,000,000 dollar"
}
],
"namespace": "securities",
"template": "SECURITY_UP"
},
{
"fields": [
{
"name": "entity_up",
"positions": [
{
"end": 959,
"start": 950
}
],
"value": "SenseTime Group"
}
],
"namespace": "securities",
"template": "SECURITY_UP"
}
]