KR102217886B1 - Exploration System and Method of Optimal Weight of Big Data-based Commodity Investment Recommendation Algorithm Using Artificial Intelligence - Google Patents

Abstract

The present invention constructs big data for products of a common series such as real estate, stock products, funds, bonds, lotto, sports Toto, etc., and calculates an optimal weight set from which the best hit rate is derived using artificial intelligence, As a technology for a system that recommends products with high investment value among products, the factor evaluation unit calculates the evaluation value of the element data for each product, and the summation score is summed by applying weights to the evaluation values of each element data for each product. It calculates and calculates the hit rate of the expected profit ranking by using the prediction ranking calculation unit that sets the expected profit ranking of the product in the order of the highest combined score, and the actual profit ranking and expected profit ranking for each product, and calculates the best hit rate for the highest hit rate. And a weight calculator for deriving a weight set of.

Description

{Exploration System and Method of Optimal Weight of Big Data-based Commodity Investment Recommendation Algorithm Using Artificial Intelligence}

The present invention relates to a system and method for searching for an optimal weight of a big data-based product investment recommendation algorithm using artificial intelligence, and in more detail, big data on products of a common series such as real estate, stock products, funds, and bonds. It is a technology related to a system that recommends products with high investment value among a number of products by constructing and calculating the optimal weight set from which the best hit rate is derived using artificial intelligence.

Various analysis techniques have been proposed to select products with promising future profits among products that can be predicted and invested, such as real estate, stock products, funds, bonds, lotto, and sports Toto.

For example, conventional parametric models such as linear regression models have been used to estimate or determine real estate prices. However, the parametric model has a problem that the reliability of the estimated price is limited because there are many strict assumptions such as the normality of the data, the independence of the explanatory variable, and the linearity of the price function. Real estate prices have various nonlinear factors that cannot be determined only by parameter models, but the existing models cannot reflect them. For example, as the land area increases, the transaction unit price tends to decrease, but the parameter model could not reflect this characteristic due to the linearity assumption. In addition, the real estate price varies greatly depending on its physical location, which is called spatial dependence. In the past, the characteristics of real estate price formation, that is, elements of physical location or distance, were not reflected, and thus the accuracy of real estate price estimation was inevitably lowered.

As described above, a system that predicts the future value of products such as real estate, stock products, funds, bonds, lotto, and sports Toto requires an evaluation system to verify its effectiveness, but an investment recommendation system equipped with such an evaluation system. Is not much.

Furthermore, it is obvious that the appropriate weight for each element data can be reflected when determining the future value of each product only when the factor data necessary to determine the future value of a product actually has an influence on the future value determination and the magnitude of the influence. However, the system with these standards is insufficient.

Korean Registered Patent Publication 10-1794027

Accordingly, the present invention has been proposed to solve the problems of the prior art, and the object of the present invention is to construct big data on products of a common series such as real estate, stock products, funds, and bonds, and use artificial intelligence. Thus, by calculating an optimal weight set from which the best hit rate is derived, it is to provide a system for recommending a product having a high investment value among a number of products.

In order to achieve the above object, the system for searching the optimal weight of the big data-based product investment recommendation algorithm according to the technical idea of the present invention using artificial intelligence includes an element evaluation unit that calculates an evaluation value of element data for each product; ; A predicted ranking calculator configured to calculate a sum score by applying a weight to an evaluation value of each element data for each product and then adding the sum, and setting an expected profit ranking of the product in the order of the highest sum score; And a weight calculating unit that calculates a hit rate of the expected return ranking using the ranking of the actual rate of return for each product and the expected rate of return, and derives an optimal weight set for which the highest hit rate is calculated.

In addition, the weight calculator may set a value within a preset range for each element data as a weight, and a sum of weights included in the weight set may be the same as a maximum value among the preset ranges.

)은 레퍼런스 데이터가 존재하는 모든 연도 개수(r), 상품의 개수(n), 실제 수익률 순위(

), 상품의 요소데이터의 개수(a), 기 설정된 범위 및 기준값을 만족하는 가중치 세트의 경우의 수(M), 요소데이터별 평가값(

) 및 레퍼런스 데이터의 요소데이터별 가중치(

)를Also, the highest hit rate (

) Is the number of all years in which reference data exists (r), the number of products (n), and the actual return rank (

), the number of product element data (a), the number of weight sets that satisfy the preset range and reference value (M), the evaluation value for each element data (

) And the weight of each element data of the reference data (

)

It may be characterized by substituting for and calculating.

In addition, the weight calculator may be characterized in that it searches for the optimal weight set using artificial intelligence.

), 상품의 요소데이터의 개수(a), 요소데이터별 평가값(

) 및 요소데이터별 가중치(

)를In addition, the weight calculation unit includes a hit rate calculator that calculates a hit rate of the expected earnings ranking by using the ranking of the actual rate of return for each product and the expected rate of return, and the hit rate (HR, Hit ratio) is the number of products (n ), the actual return rank (

), number of product element data (a), evaluation value for each element data (

) And weight by element data (

)

It may be characterized by substituting for and calculating.

On the other hand, in order to achieve the above object, the method of searching for the optimal weight of the big data-based product investment recommendation algorithm according to the technical idea of the present invention using artificial intelligence is: (a) The input unit retrieves reference data corresponding to the product. Receiving an input, (b) setting a value within a preset range for each element data as a weight, and (c) a hit ratio calculator, a prediction calculated using the element data included in the reference data and the weight The step of calculating the hit rate of the expected profit ranking by using the profit ranking and the actual profit ranking for each product, and (d) when the hit rate is the highest hit rate, the weight calculation unit optimizes the weight set used to calculate the highest hit rate. It characterized in that it comprises the step of determining the weight set.

In addition, when the hit rate is not the highest hit rate in step (d), the process moves to step (b), and in this case, step (b) may be characterized in that a different weight set is set for the element data.

), 상품의 요소데이터의 개수(a), 요소데이터별 평가값(

) 및 요소데이터별 가중치(

)를In addition, the hit ratio (HR, Hit ratio) is the number of products (n), the actual profit ranking (

), number of product element data (a), evaluation value for each element data (

) And weight by element data (

)

It may be characterized by substituting for and calculating.

According to the system and method for searching for the optimal weight of the big data-based product investment recommendation algorithm according to the present invention using artificial intelligence,

First, since the products that are expected to have high returns are displayed in order among a number of products, a higher rate of return can be expected when investing.

Second, since the optimal weight is searched using the hit ratio calculation algorithm, it is possible to derive the best expected profit ranking corresponding to the difference in product classification or sub-classification.

Third, when applying the real estate evaluation of the present invention, new indicators such as energy demand and location value of jeonse energy that can be used as element data were proposed, and the reliability of the expected revenue ranking was maximized by verifying the influence of the new indicator on the hit rate.

1 is a block diagram of a system (investment recommendation system) for searching for an optimal weight of a big data-based product investment recommendation algorithm according to an embodiment of the present invention using artificial intelligence.
FIG. 2 is a diagram illustrating the display of element data for each product, expected return ranking, actual rate of return, and hit rate in a system that searches for an optimal weight of a product investment recommendation algorithm based on big data according to an embodiment of the present invention using artificial intelligence. Example drawing.
3 is a graph showing the energy demand in the Seoul region and the Gyeonggi region in the period 1999-2020 in an embodiment using the present invention for real estate evaluation.
4 is a graph showing the trading index and the jeonse index, and jeonse energy in the period 1999-2018 in the period of 1999-2018, in an embodiment using the present invention for real estate evaluation.
Figure 5 is an exemplary diagram showing that in an embodiment in which the present invention is used for real estate evaluation, element data for each real estate, expected income ranking, actual rate of return, and hit rate are displayed.
6 is a flowchart of a method of searching for an optimal weight set of a big data-based product investment recommendation algorithm according to an embodiment of the present invention using artificial intelligence.
7 is a flow chart showing a detailed process of step S110.
8 is a flow chart showing a detailed process of step S114.
9 is a case in which an investment is made in the highest ranking (top 30) among the expected profit rankings derived using the investment recommendation algorithm according to an embodiment of the present invention, 2 years later, the actual rate of return, the overall average rate of return, and the jeonse price rate are 70% or more. A graph comparing the rate of return of apartments.

With reference to the accompanying drawings, a system and method for searching for an optimal weight of a big data-based product investment recommendation algorithm according to embodiments of the present invention using artificial intelligence will be described in detail. Since the present invention can apply various changes and have various forms, specific embodiments will be illustrated in the drawings and described in detail in the text. However, this is not intended to limit the present invention to a specific form of disclosure, and it should be understood that all changes, equivalents, and substitutes included in the spirit and scope of the present invention are included. In describing each drawing, similar reference numerals have been used for similar elements.

In addition, unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by those of ordinary skill in the art to which the present invention belongs. Terms as defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and should not be interpreted as an ideal or excessively formal meaning unless explicitly defined in this application. Does not.

The system and method for searching for the optimal weight of the big data-based product investment recommendation algorithm according to an embodiment of the present invention using artificial intelligence will be implemented as including a main configuration or function in any one computer device or distributed computing system. I can.

The system and method for searching for the optimal weight of the big data-based product investment recommendation algorithm according to an embodiment of the present invention using artificial intelligence is expected to be included in the investment recommendation system 100 for calculating the expected return ranking of the product. Evaluate the hit rate of the profit ranking.

1 and 2, a system for evaluating a big data-based product investment recommendation algorithm using artificial intelligence according to an embodiment of the present invention includes an element evaluation unit 110 that calculates an evaluation value of element data for each product, and , An expected ranking calculation unit 120 configured to calculate a total score by applying a weight to the evaluation value of each element data for each product and then adding it, and setting the expected profit ranking of the product in the order in which the sum score is large, and the And a weight calculator 140 that calculates a hit ratio of the expected profit ranking by using the actual profit ranking for each product and the expected profit rank, and derives an optimal weight set for which the highest hit ratio is calculated.

The weight calculation unit 140 includes a hit ratio calculation unit 130 that calculates a hit ratio of the expected profit ranking by using the actual profit ranking for each product and the expected profit ranking.

Products include all tangible and intangible properties that can be traded, such as real estate, stock products, funds, and bonds, and products that require prediction, such as lotto and sports toto.

The name of the product and element data constituting the characteristics of each product are input to the system through the input unit 160.

The inputted product element data is stored in the storage unit 180. Since element data includes vast amounts of data such as trading prices accumulated over several decades for each product, element data can be regarded as big data related to products.

In addition, information such as an evaluation value of the element data for each product, an expected profit ranking, an actual profit ranking, and a hit rate may be output using the display unit 170. A known information output device such as a monitor or a printer may be applied to the display unit 170.

It is preferable that the investment recommendation system 100 calculates an expected profit ranking for one common product. One common series may be, for example, an apartment (real estate) located in a specific area, a stock product corresponding to the same industry, a fund that invests in similar products, and the like.

The hit ratio calculation unit 130 that calculates the hit ratio of the expected profit rankings for each product may use an artificial intelligence algorithm for rapid analysis of vast amounts of data.

Below, the present invention will be described using an embodiment in which the investment recommendation system 100 is applied to the real estate field.

Real estate (product) is a category of real estate such as land, housing (apartment), etc., and includes objects that can be sold.

Each real estate includes element data in detail. The factor data may include the equilibrium of apartment complexes, the jeonse price rate, the investment gap (the difference between the selling price and the jeonse price), the year (open or occupied year), the number of households, the rate of increase in the sale price for 6 months, and the rate of increase in the jeonse price for one year.

The element data can include energy demand. Demand energy is an index that can be used to determine whether the supply of apartment households in an area is appropriate. The energy demand is the cumulative sum of the difference between the amount of demand and the amount of supply. If the amount of supply is large, the value of energy demand decreases, and if the amount of supply is small, the value of energy demand increases. At this time, the amount of demand is calculated as the product of the number of people and the life cycle of the household. The value of the household life cycle may be 0.5%, which is usually a value calculated as an appropriate amount of demand per year. If you want to calculate the quantity demanded by will, divide this value by 12. The amount of supply is calculated when real estate (housing) is completed in the area and it is possible to move in.

3(a) is a graph showing energy demand in the Seoul area. This graph shows the flow of energy demand calculated from the cumulative sum of the difference between the quantity of demand and supply each month since 1998. From 2005 to 2016, energy demand continued to increase due to insufficient supply, but from 2016, the amount of supply increased significantly, showing a trend in which energy demand remained in place.

3(b) is a graph showing energy demand in Gyeonggi Province. In Gyeonggi Province, energy demand rose from 2005 to the end of 2016 due to supply shortages, but since then, supply has increased and has declined until around 2020. However, after 2020, energy demand is rising again due to the absence of additional supply. At this time, it can also be seen that the energy demand in 2020 is about the level of May 2013.

The higher the energy demand is, the less the supply is, so the possibility of an increase in the trading price is higher, and the lower the energy is, the less the possibility of an increase in the sale price.

Demand energy shows a relatively higher impact on the hit rate when evaluating real estate outside the metropolitan area than on real estate in the metropolitan area. Therefore, energy demand can be viewed as relatively more important element data when evaluating real estate outside the metropolitan area.

Demand energy can be applied not only to the city/province unit, but also to the sub-unit, ward/city unit.

Demand energy can be used as an important measure not only of the current position value, but also the rate of increase (slope). The hit ratio calculated after applying the weight as the largest value to the current position value or the rising rate of the demand energy is shown in Table 1.

Time Hit rate of current position value (%) 200801 60.84 200901 62.12 201001 68.48 201101 58.21 201201 72.15 201301 60.67 201401 66.45 201501 51.72 201601 59.26 Average 62.21

In addition, the element data may include the location value of the jeonse energy. Fig. 4(a) is a graph of the chartered energy in Seoul, which is calculated as the cumulative sum of the difference between the increase in jeonse and the increase in jeonse in Seoul.

The location value of jeonse energy is the cumulative flow of the difference between the rise in jeonse price and the increase in trading price. In addition, the current location of the jeonse energy means the level of the current location compared to the maximum value of the jeonse energy. The current location of the jeonse energy can be expressed as follows by an equation.

[Equation 1]

는 현재 전세에너지,

는 최대 전세에너지,

는 최저 전세에너지이다.

Is the current jeonse energy,

Is the maximum jeonse energy,

Is the lowest jeonse energy.

Real demand is the demand to live in the form of purchase or charter for the purpose of residence without considering future value increase. On the other hand, demand is a form of purchase in which investment is made for the purpose of increasing future value other than the purpose of living. Jeonse demand, singer demand, and real demand residing in jeonse appear in a tug of war.

The graph of the jeonse energy visually shows the demand and real demand. In Seoul, after a mistaken request for jeonse energy to rise, it leads a bullish market by declining and discharging jeonse energy, and then the trading index is also showing a pattern of declining when jeonse energy falls too much. On the other hand, in local markets outside of the metropolitan area, where the flow is strictly focused on real demand, jeonse energy rarely falls below zero and then rises.

The location value of jeonse energy shows a relatively higher impact on the hit rate when evaluating real estate in the metropolitan area than on real estate outside the metropolitan area. Therefore, the location value of jeonse energy can be included in the element data only when evaluating real estate in the metropolitan area.

The hit ratio calculated after applying the weight as the largest value to the jeonse energy position value is shown in Table 2.

Time Of the location value of the jeonse energy
Hit rate (%) 200801 71.64 200901 80.73 201001 84.53 201101 83.36 201201 68.49 201301 73.02 201401 70.11 201501 70.07 201601 79.41 Average 75.71

When the weight of the location value of jeonse energy is large, a higher hit rate appears than that of energy demand, so it can be seen that the location value of jeonse energy is more important than the energy demand in real estate evaluation. Therefore, when deriving the expected profit ranking, the jeonse energy potential value is given a greater weight than the energy demand.

In addition, the factor data is data that measures the possibility of local increase in real estate, including household growth rate, jeonse price rate, unsold sales, unsold sales after completion, trading volume compared to last year, and neighboring occupancy after two years.

The household growth rate is obtained by calculating how much the number of households in the region has increased compared to last year by using the number of households by year of the district provided by the National Statistical Office.

The jeonse price rate is the ratio of the jeonse price to the current trading price. It is the ratio of the average sale price and jeonse price for each apartment, and the number is calculated as the average value for the same unit and the average value for the region (city unit).

On the other hand, the neighborhood location of the jeonse price rate indicates which real estate occupies and how much position in the regional average jeonse price rate. In this case, the region refers to the unit of eup/myeon/dong. The equation for calculating the neighborhood location of the jeonse price rate is as follows.

[Equation 2]

는 현재 부동산의 전세가율,

는 지역 최대 전세가율,

는 지역 최저 전세가율이다. 일반적으로 시·구 단위의 지역은 전세가율은 높을수록 유리하다.

Is the current real estate jeonse rate,

Is the local maximum jeonse rate,

Is the region's lowest jeonse rate. In general, the higher the jeonse rate is, the more advantageous for a city/gu-level area.

Unsold and unsold after construction are data provided by Statistics Nuri of the Ministry of Land, Infrastructure and Transport.

The volume compared to last year is the rate of increase in volume compared to the same month last year in the region.

The amount of occupancy in the vicinity after two years refers to the amount of occupancy in the region and the neighboring city/gu level at the time the jeonse expires two years later.

[Table 3] and [Table 4] show the results of calculating the hit ratio by putting a weight of 100 on any one of the household growth rate, jeonse rate, unsold, unsold after completion, trading volume compared to last year, and neighboring occupied volumes two years later.

Time Hit rate of household growth rate (%) Hit rate of trading volume compared to last year Hit rate of neighboring occupants after 2 years (%) 200801 61.05 65 61.74 200901 54.24 54.6 59.25 201001 56.94 59.78 65.04 201101 60.62 49.02 59.45 201201 63.7 61.07 58.64 201301 67.91 49.59 52.96 201401 70.84 62.94 48.79 201501 56.78 52.15 57.1 201601 58.47 56.03 55.78 Average 61.17 56.69 57.64

Time Unsold
Hit rate (%) Hit rate of unsold sales after completion (%) Hit rate of the jeonse rate (%) 200801 57.28 57.76 69.36 200901 61.26 61.27 75.96 201001 66.3 64.31 88.26 201101 64.72 74.82 91.01 201201 55.75 66.35 81.77 201301 65.71 63.85 86.28 201401 67.69 70.25 75.62 201501 66.69 60.87 68.39 201601 64.99 65.38 72.82 Average 63.38 64.98 78.83

The jeonse price rate was found to have a higher overall hit rate than other element data. The year 2015 was also low compared to other periods, but it was very high in 2010 and 2011.

The predicted ranking calculation unit 120 applies a weight to the evaluation value of each element data for each real estate, and then calculates a sum score by summing the evaluation values to which the weight is applied, and the expected profit ranking of the real estate in the order of the highest sum score. Is set.

[Table 5] shows that in applying the weights included in the predetermined range to the evaluation value of each element data, the weight calculation unit 140 obtains a weight that can be commonly applied to element data of all years of a specified product for each element data. After the hit ratio was calculated. It can be seen that if the optimal weight set is applied to the specified product group, excellent hit ratio results can be obtained.

Time Hit rate (%) 200801 81.53 200901 85.07 201001 88.80 201101 88.61 201201 83.59 201301 86.65 201401 81.50 201501 82.87 201601 85.21 Average 84.87

The influence test unit 150 numerically shows how much influence each element data has by using an algorithm of a hit rate.

In addition, by calculating all the number of weight cases of each element data using the weight calculator 140, an optimal weight set for deriving the highest hit rate may be obtained. That is, an embodiment of the present invention can be used as a tool for predicting a region or real estate with a high possibility of rising when using the pattern and weight of past element data of a product.

The product includes a number of element data, and a weight is set for each element data. Therefore, the weight set means a weight group of a number corresponding to the number of element data.

In detail, the element data that determines the real estate with high probability of increase in the region include the apartment complex equilibrium, jeonse price rate, investment gap (difference between sales price and jeonse price), year (open or occupied year), number of households, and sales price increase rate for 6 months, The rate of increase in jeonse prices for one year is included.

Referring to FIG. 5, the element evaluation unit 110 may evaluate element data for each real estate to calculate an evaluation value and then list it in a table format.

The evaluation value becomes a value calculated through a preset operation method for each element data.

The optimal weight set is a value from which the highest hit rate can be derived, and the weight calculator 140 searches for it.

The expected ranking calculation unit 120 may calculate the summed score in the same manner as in Equation 4.

[Equation 4]

Estimated profit rankings are set in the order of the highest total score as the number of products.

The actual rate of return is calculated by using the product price at the time when the expected ranking calculation unit 120 sets the expected return ranking and the product price after a certain period of time. After a certain period of time, it may be randomly selected from 1 day to 10 years, or it may be a time point at which the price of a product can be calculated in the future, or after that. The hit rate calculation unit 130 sets the rankings of the products in the order of the highest actual rate of return. The actual rate of return may be the rate of change in price from the point when the previous expected rate of return is set up to the point of evaluation, but if there are variables such as an apartment complex where the size of each household is different and various price distributions are formed within one apartment complex. Data, such as the sum of households with rising prices, can also be applied to the actual rate of return.

The hit rate calculation unit 130 calculates a hit rate by comparing the expected profit ranking and the actual profit ranking. Specifically, the hit ratio calculation unit 130 includes a process of deriving a difference value by subtracting the actual profit ranking for each product from the expected profit ranking for each product, and calculating an actual sum value obtained by summing the difference values of the products. The difference value is expressed as an absolute value. [Table 6] and [Table 7] are examples of calculating the difference and sum values when the number of products is 100. [Table 6] shows the cases where the actual and expected earnings rankings are exactly the same, and [Table 7] shows the cases where the actual and expected earnings rankings are completely different.

Estimated profit ranking Actual revenue ranking Difference One One 0 2 2 0 3 3 0 … … 0 98 98 0 99 99 0 100 100 0 Sum 0

Estimated profit ranking Actual revenue ranking Difference value One 100 99 2 99 97 3 98 95 … … 98 3 95 99 2 97 100 One 99 Worst expected sum value 5000

Referring to [Table 6], if the actual and expected profit rankings are perfectly the same, the sum is 0.

Referring to [Table 7], if the difference between the actual and expected profit rankings is completely different and the difference is maximized, the sum is 5000.

The difference value is expressed as an absolute value, and the larger the difference value, the different the actual and expected income ranking.

The maximum value of the difference and sum is determined by the number of products.

The hit ratio calculating unit 130 calculates the hit ratio of the expected profit ranking by using the worst predicted sum value and the actual sum value calculated on the assumption that the difference value for each product is the largest.

The worst expected sum value (BEV) is calculated by substituting the number of difference values (the number of products) (n) into [Equation 5].

[Equation 5]

은 실제 수익 순위, a는 상품의 요소데이터의 개수,

는 요소데이터별 평가값,

는 요소데이터별 가중치이다.The hit ratio (HR, hit ratio) is calculated by substituting the worst predicted sum value (BEV) and the actual sum value into [Equation 6]. In this case, n is the number of products,

Is the actual profit ranking, a is the number of element data of the product,

Is the evaluation value for each element data,

Is the weight for each element data.

[Equation 6]

[Table 8] shows that the number of products is 100, and the product with the expected profit ranking of 1 is the 100th in the actual rate of return, and the 2nd to 100th of the expected profit ranking has a difference of 1 from the actual profit ranking. It is the summation value in case.

Estimated profit ranking Actual revenue ranking Difference One 100 99 2 One One 3 2 One … … … 98 97 One 99 98 One 100 99 One Actual sum value 198

Substituting the first case into [Equation 6], the hit rate is 96.04%.

[Table 9] shows that the number of products is 100, and the products with the 1st and 2nd expected profit rankings are ranked 100th and 99th respectively in the actual profit ranking, and the difference between the 3rd and 100th expected profit rankings from the actual profit ranking. Is the sum value in the second hypothetical case.

Estimated profit ranking Actual revenue ranking Difference value One 100 99 2 99 97 3 One 2 … … … 98 96 2 99 97 2 100 98 2 Actual sum value 390

Substituting the second case into [Equation 6], the hit rate is 92.2%.

The weight calculator 140 sets a value of a preset range for each element data as a weight. In particular, it is conditional that the sum of weights included in the weight set is always equal to the reference value. For example, if the preset range is 0 to 100, element data is 5, and the reference value is 100, the weight of element data 1 is 10, the weight of element data 2 is 20, the weight of element data 3 is 30, The weight of element data 4 is 15, and the weight of element data 5 is 25, so the total sum of the five weights is always set equal to the reference value.

In order to search for the optimal weight set from which the highest hit rate is derived, reference data is input to the system, the weight calculator 140 sets a weight set for the reference data, and the predicted ranking calculation unit 120 ranks the predicted profit. And the process of calculating the hit ratio in comparison with the actual profit ranking by the hit ratio calculator 130 may be repeated.

The reference data may be information on past element data of a corresponding product, information on element data of a similar product, and the like. For example, the historical element data information of the corresponding product may be element data from 2000 to 2017 of the corresponding apartment in order to calculate the 2018 expected profit ranking of apartment A to apartment Z. In addition, the element data information of similar products may be element data of Company B, which has been listed on the KOSDAQ in the past to calculate the expected profit ranking of Company A newly listed on the KOSDAQ and has similar business types and sizes.

The greater the number of element data and the wider the preset range is, the higher the derived maximum hit rate may be, but the number of combinations of weights in the weight set also increases exponentially. That is, it is preferable that the search for a meaningful optimal weight set is performed automatically because it requires considerable computation time. In order to shorten the time to search for an optimal weight set, the weight calculation unit 140 includes a calculation function by artificial intelligence.

)은 레퍼런스 데이터가 존재하는 모든 연도 개수(r), 상품의 개수(n), 상품의 요소데이터의 개수(a), 기 설정된 범위 및 기준값을 만족하는 가중치 세트의 경우의 수(M), 레퍼런스 데이터의 요소데이터별 가중치(

)를 [수학식 7]에 대입하여 산출할 수 있다.Best hit rate(

) Is the number of all years in which reference data exists (r), the number of products (n), the number of element data of the product (a), the number of weight sets satisfying the preset range and reference value (M), the reference Weight of each element of data (

) Can be calculated by substituting in [Equation 7].

[Equation 7]

)이 산출될 때의 요소데이터별 가중치(

)들이 최적의 가중치 세트가 된다.Best hit rate(

The weight for each element data when) is calculated (

) Are the optimal weight set.

Element data is different for the classification of products, i.e. real estate, stock items, and fund products, but the sub-classification of the product, for example, if the classification is real estate, the metropolitan area and regions outside the metropolitan area, and each city/province The type or influence of the element data is also different depending on the difference in sub-classifications of and eup/myeon/dong units. Therefore, the optimal weight set is different from each other according to the classification and subdivision of the product. The weight calculator 140 searches for an optimal weight set according to each situation.

In addition, an embodiment of the present invention replaces the weight applied to the selected element data with the largest value or the smallest value among the number of preset ranges, and analyzes the influence of the selected element data using the varying degree of hit rate. It may further include an influence test unit 150.

When the highest or lowest weight is arbitrarily assigned to the selected element data to examine the influence of the element data, the calculated hit rate differs from the highest hit rate, and the influence test unit 150 analyzes the difference and Shows the influence of.

Next, a method of searching for an optimal weight set of a big data-based product investment recommendation algorithm according to an embodiment of the present invention will be described.

Referring to FIG. 6, in the big data-based product investment recommendation method, the weight calculation unit 140 calculates all the number of weight cases of each element data using reference data corresponding to the product. A step of calculating a set (S110), an input unit 160 receiving element data for each product (S120), and an element evaluation unit 110 calculating an evaluation value of the element data for each product (S140), and , The predicted ranking calculation unit 120 calculating a sum score by applying an optimal weight to each of the element data for each product and then adding them, and setting the expected profit ranking of the product in the order of the greatest sum score (S160) )Wow; The display unit 170 includes a step (S180) of displaying the expected profit ranking for each product.

The formula for calculating the highest hit rate in step S110 is as shown in [Equation 7].

Specifically, referring to FIG. 7, in step S110, the input unit 160 receives reference data corresponding to the product (S110), and the weight calculator 140 sets a value within a preset range for each element data as a weight. Step (S112), and the hit ratio calculating unit 130 calculating the hit ratio of the expected profit ranking using the expected profit ranking calculated using the element data included in the reference data and the weight and the actual profit ranking for each product. It includes (S114).

The weight calculation unit 140 determines whether the hit ratio calculated in step S114 is the highest hit ratio (S116). If the hit ratio calculated in step S114 is the highest hit ratio, the weight calculator 140 determines the weight set used to calculate the highest hit ratio as the optimal weight set (S118). However, if the hit ratio calculated in step S114 is not the highest hit ratio (S116), the process moves to step S112 and the weight calculator 140 sets a different weight set.

The calculation equation for calculating the hit rate in step S114 is as shown in [Equation 6].

Referring to FIG. 8, in step S114, the element evaluation unit 110 calculates an evaluation value of element data based on the reference data for each product (S1141), and the expected ranking calculation unit 120 calculates each element data for each product. The step of calculating a sum score by adding a weight to and then adding the sum, and setting the expected profit ranking of the product in the order of the highest sum score (S1142), and the hit ratio calculating unit 130 is the actual profit ranking for each product and the expected And calculating the hit rate of the expected profit ranking by using the profit ranking (S1144 to S1148).

In the step of calculating the hit rate of the expected profit ranking by using the actual profit ranking for each product and the expected profit ranking by the hit rate calculating unit 130, the difference value is derived by subtracting the actual profit ranking for each product from the expected profit ranking for each product. Step (S1144), calculating an actual sum value obtained by summing the difference values of the products (S1145), and calculating the worst expected sum value on the assumption that the difference value for each product is the largest (S1146), the And calculating a hit ratio of the expected profit ranking by using the actual sum value and the worst expected sum value (S1148).

The worst predicted sum value calculated in step S1146 is calculated by substituting the number n of difference values into [Equation 5].

[simulation]

A rate of return was calculated when the investment recommendation algorithm according to an embodiment of the present invention was applied to the real estate (apartment) field.

For information that contrasts with the return of the investment recommendation algorithm, the overall average rate of return and the rate of return of apartments with a jeonse rate of 70% or more were used.

Using a large number of factor data and optimal weight set, the predicted profit rankings were derived for each apartment for a period of one month for a period of 10 years, and the highest hit rate was calculated by comparing the actual profit rankings after two years. The result of the calculated best hit rate is shown in [Table 10].

Time Best Hit Rate (%) 200801 81.53 200901 85.07 201001 88.80 201101 88.61 201201 83.59 201301 86.65 201401 81.50 201501 82.87 201601 85.21 Average 84.87

9 shows an actual rate of return, an overall average rate of return, and a jeonse price rate of 70% or more after 2 years when investing in the highest ranking (top 30) among the expected return rankings derived using the investment recommendation algorithm according to an embodiment of the present invention. This is a graph comparing the rate of return of apartments. As shown in the graph, it can be seen that the return on investment can be greatly improved when an apartment to be purchased is selected using an investment recommendation algorithm according to an embodiment of the present invention.

Although a preferred embodiment of the present invention has been described above, the present invention can use various changes, modifications, and equivalents. It is clear that the present invention can be applied in the same manner by appropriately modifying the above embodiments. Therefore, the above description does not limit the scope of the present invention determined by the limits of the following claims.

100: investment recommendation system 110: element evaluation department
120: predicted ranking calculation unit 130: hit ratio calculation unit
140: weight calculation unit 150: influence test unit
160: input unit 170: display unit
180: storage unit

Claims (9)

An element evaluation unit for calculating evaluation values of element data for each product;
A predicted ranking calculator configured to calculate a sum score by applying a weight to an evaluation value of each element data for each product and then adding the sum, and setting an expected profit ranking of the product in the order of the highest sum score;
Comprising a weight calculator for calculating a hit ratio of the expected profit ranking using the actual profit ranking for each product and the expected profit ranking, and deriving an optimal weight set for calculating the highest hit ratio,
The weight calculation unit
And a hit ratio calculating unit that calculates a hit ratio of the expected profit ranking by using the actual profit ranking for each product and the expected profit ranking,
The hit ratio (HR, Hit ratio) is the number of products (n), the actual profit ranking (

), number of product element data (a), evaluation value for each element data (

) And weight by element data (

)

A system for searching for an optimal weight of a product investment recommendation algorithm, characterized in that calculated by substituting in. The method of claim 1,
The weights included in the weight set are set among values within a preset range, and the sum of the weights included in the weight set is always equal to a reference value. The method of claim 2,
The highest hit rate above (

) Is the number of all years in which reference data exists (r), the number of products (n), and the actual profit ranking (

), the number of product element data (a), the number of weight sets satisfying the preset range and reference value (M), the evaluation value for each element data (

) And the weight of each element data of the reference data (

)

A system for searching for an optimal weight of a product investment recommendation algorithm, characterized in that calculated by substituting in. The method of claim 2,
The weight calculation unit searches for the optimal weight set of the product investment recommendation algorithm by using artificial intelligence. delete Calculating an optimal weight set from which the highest hit rate is derived by using the reference data corresponding to the product by a weight calculator;
The step of receiving element data for each product by the input unit, and
The step of calculating the evaluation value of the element data for each product by the element evaluation unit,
Comprising the step of calculating a sum score by applying an optimal weight set to the element data for each product, and then adding the predicted ranking calculation unit, and setting the expected profit ranking of the product in the order of the greatest sum score,
The weight calculation unit
And a hit ratio calculating unit that calculates a hit ratio of the expected profit ranking by using the actual profit ranking for each product and the expected profit ranking,
The hit ratio (HR, Hit ratio) is the number of products (n), the actual profit ranking (

), number of product element data (a), evaluation value for each element data (

) And weight by element data (

)

A method of searching for an optimal weight of a product investment recommendation algorithm, characterized in that calculated by substituting in. (a) receiving reference data corresponding to the product by the input unit; and
(b) the step of setting, by the weight calculation unit, a value within a preset range for each element data as a weight,
(c) calculating, by a hit rate calculating unit, a hit rate of the expected profit ranking using the estimated profit ranking calculated using the element data included in the reference data and the weight, and the actual profit ranking for each product;
(d) when the hit ratio is the highest hit ratio, including the step of determining, by a weight calculator, a weight set used to calculate the highest hit ratio as an optimal weight set,
The hit ratio (HR, Hit ratio) is the number of products (n), the actual profit ranking (

), number of product element data (a), evaluation value for each element data (

) And weight by element data (

)

A method of searching for an optimal weight of a product investment recommendation algorithm, characterized in that calculated by substituting in. The method of claim 7,
If the hit ratio in step (d) is not the highest hit ratio, it moves to step (b),
In this case, in step (b), a different weight set is set for the element data. delete

Images