KR101679249B1 - Analysis method and device for scientific research information revolution - Google Patents

Abstract

Collecting at least two scientific research documents in a predetermined field through a collection unit; Preprocessing the scientific research literature through a preprocessing unit and constructing an authorship evolution model of two or more scientific research literature; And acquiring the scientific research information evolution result by setting the scientific research information estimation parameter through the acquisition unit and calculating the scientific research information estimation parameter by using the evolutionary operation formula. Lt; / RTI > In the embodiment provided by the present invention, the scientific research literature is collected in a predetermined field, the preprocessing is carried out on the scientific research literature, the authors subject evolution model of the scientific research literature is constructed, In order to obtain the results of the evolution of research information, the results of the evolution of scientific research information should include the topics of the scientific research literature and the research interests of the researchers, as well as the temporal change rules such as the scientific research theme of the pre- It is advantageous to grasp the evolutionary context of the scientific research subject in the predetermined field.

Description

Technical Field [0002] The present invention relates to a method and apparatus for analyzing the evolution of scientific information,

The present invention relates to the field of scientific research information analysis technology, and more particularly, to a method and apparatus for analyzing scientific research information evolution.

Scientific research literature is the main storage medium of scholarly achievement. It is a mass of sweat and wisdom of researchers. It is an important means of knowledge dissemination and academic exchanges in human society. Lt; / RTI > The Price Science Index and the Logic Curve Growth Model show that the quantity of scientific research literature is increased to an exponential order. In order to accurately grasp the state of science and technology development, it is possible to combine the feature information of scientific research literature and find out the science technology subject and its internal development context from the vast scientific research literature data. Among them, scientific research literature resources are characterized by a large amount of feature information, for example, the relationship of potential words between words and words, the relationship between the subject of the scientific research literature and the author (research interests of the author) And the like.

In the prior art, Rosen-Zvi et al. Introduced the authors' hidden variables into the LDA model and replaced the scientific literature-subject distribution in the LDA model with an Author-Topic (AT) distribution. The AT model can efficiently identify the relationship between the author of the scientific research literature and the subject of the scientific research literature, and can find the research interest of the researcher. However, in order to overcome these limitations, Mimno et al. Proposed an Author-Persona-Topic (APT) model on the basis of an AT model. . These models correspond to the 'persona' of the researcher and the research interest, and provide an enlightenment method for estimating the amount of research interest. The AT and APT models, however, did not directly consider the effects of time factors when considering the research interests of the researchers, only considering the scientific research literature, Can not post about the temporal change rule of.

The present invention provides a method and apparatus for analyzing the evolution of scientific research information for solving a problem that can not be posted on the subject of scientific research literature and time change rules of research interest of researchers in the prior art.

In order to solve the above problems in the prior art, the present invention relates to a method for collecting scientific research documents in a field set in advance through a collection unit; Preprocessing the scientific research literature through a preprocessing unit and building an authorship evolution model of the two or more scientific research literature; And acquiring the scientific research information evolution result by setting the scientific research information estimation parameter through the acquisition unit and calculating the scientific research estimation parameter by using the evolutionary operation formula, thereby providing an analysis method of scientific research information evolution do.

The invention also relates to an apparatus for analyzing the evolution of scientific research information, comprising: a collection unit, a preprocessing unit and an acquisition unit, the collection unit collecting two or more scientific research articles in a predetermined field; Wherein the preprocessing unit preprocesses the scientific research literature and builds an authorship evolution model of the two or more scientific research literature; The acquisition unit further provides a device for analyzing scientific research information evolution that acquires scientific research information evolution results by setting scientific research information estimation parameters and computing the scientific research estimation parameters using an evolutionary operation formula.

The embodiments provided by the present invention provide the following beneficial effects.

In this embodiment, two or more scientific research documents are collected in a predetermined field, preprocessing is carried out on scientific research documents, an authors subject evolution model of two or more scientific research documents is constructed, and a scientific research information estimation parameter The results of the evolution of scientific research information are obtained by posting the research topics of the scientific research literature and the research interests of the researchers, It is advantageous to grasp the evolutionary context of science research subject in pre-set field because it is possible to find and publish time change rule such as scientific research theme of scientific research field and research interest of researcher.

The foregoing and / or additional aspects and advantages of the present invention will become more apparent from the following description of the embodiments thereof, taken in conjunction with the accompanying drawings, in which:
1 is a flowchart of a first embodiment of a method for analyzing scientific research information evolution according to the present invention.
2 is a flowchart of a second embodiment of a method for analyzing scientific research information evolution according to the present invention.
3 is a probability diagram of the AToT model in this embodiment.
FIG. 4 is a trend diagram of the temporal change of the theme 'reinforcement of learning' in the present embodiment.
FIG. 5 is a trend diagram of temporal change of the subject 'EM and mixed model' in the present embodiment.
FIG. 6 is a trend diagram of temporal change of the subject 'visual identification and factor analysis' in the present embodiment.
FIG. 7 is a trend diagram of the temporal change of the subject 'data model and learning algorithm' in the present embodiment.
FIG. 8 is a trend diagram of temporal change of the theme 'support vector machine and kernel technique' in this embodiment.
FIG. 9 is a trend diagram of temporal change of the subject 'neural network' in the present embodiment.
FIG. 10 is a trend diagram of temporal change of the theme 'language identification' in the present embodiment.
11 is a trend diagram of temporal change of the theme 'Bayesian learning' in this embodiment.
12 is a time distribution diagram of scientific research documents in this embodiment.
13 is a time evolution model diagram of the research subject of the present embodiment.
FIG. 14 is a schematic diagram showing a comparison of the perplexity of the AT model and the AToT model in this embodiment.
FIG. 15 is a structural diagram of a first embodiment of an apparatus for analyzing scientific research information evolution according to the present invention. FIG.
16 is a structural diagram of a second embodiment of an apparatus for analyzing scientific research information evolution according to the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings, in which like or similar reference numerals denote elements having the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the drawings are illustrative and are used merely for interpreting the present invention and are not used to limit the present invention.

Those skilled in the art will recognize that the singular forms "a," " an ", "above" and "include" Also, the phrase "comprising" as used herein should be interpreted as indicating the presence of said features, integers, steps, operations, elements and / or modules but may include one or more other features, integers, And / or does not exclude the presence or addition of these groups. When an element is referred to as being "connected" or "coupled" to another element, it should be understood that the element may be directly coupled or coupled to another element, or an intermediate element may be present. In addition, "connection" or "coupling ", as used herein, may include wireless connection or coupling. As used herein, the phrase "and / or" includes any or all combinations of one or more related enumerated terms.

It will be apparent to those skilled in the art that, except as otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs do. Also, for example, commonly used predefined terms have the meaning consistent with the meanings in the upper and lower statements of the prior art, and are not to be construed as idealized or overly formal meanings except as defined herein I must understand.

1 is a flowchart of a first embodiment of a method for analyzing scientific research information evolution according to the present invention. As shown in Fig. 1, the operation process of the analysis method of scientific research information evolution according to the present embodiment includes the following steps.

Step 101: Collect more than one scientific research literature in a predetermined field through the collection unit.

In the present embodiment, the functions of the acquisition unit, the preprocessing unit, and the acquisition unit can be performed through the computer. In this step, the collection unit collects two or more scientific research documents in a predetermined field. For example, the collection unit can collect 1000 scientific research documents in a scientific research field. After collecting two or more scientific research documents in a predetermined field through the collection unit, step 102 is entered.

Step 102: Preprocessing the scientific research literature through a preprocessing unit and constructing an authorship evolution model of two or more scientific research literature.

), 저자 주제 파라미터(

), 과학연구 문헌(m) 중 제 n개 단어의 주제 배분(

) 및 과학연구 문헌(m) 중 제 n개 단어의 저자 배분(

)을 계산하는 것을 포함한다. 그 중, 깁스 샘플링 알고리즘 공식은 아래와 같으며,In this step, all the scientific research articles collected through the preprocessing unit are preprocessed, including filtering out of use words, numbers and words lower than a predetermined frequency, and the name of the author of the scientific research document The method of resolving the name of the name includes the artificial neutralization method, the semi-automatic neutralization method, the automatic neutralization method and the mixed neutralization method. And a set of science and technology workers who are made up of A authors. In addition, the presentation time of the scientific research literature is normalized to the interval (0, 1). Author-Topic over Time (AToT) of each collected scientific research literature is constructed by using the Gibbs sampling algorithm,

), Author subject parameter (

), The theme allocation of the n words in the scientific research literature (m)

) And the authors' distribution of the n words in the scientific research literature (m)

). ≪ / RTI > Among them, the formula of Gibbs sampling algorithm is as follows,

는 주제(k)의 시간적 변화 베타(Beta) 분포 파라미터 벡터이며，

는 과학연구 문헌(m) 중 제 n개 단어의 주제 배분이고,

는 과학연구 문헌(m) 중 제 n개 단어의 저자 배분이며,

는 과학연구 문헌(m) 중 제 n개 단어이며,

는 과학연구 문헌(m) 중 제 n개 단어의 타임 스탬프이며,

는

（a = 1, …, A）의 디리클레(Dirichlet) 선험 파라미터 벡터이며,

는 저자 (a)의 주제 확률 분포 파라미터이며,

는

（k = 1, …, K）의 디리클레 선험 파라미터 벡터이며,

는 주제(k)의 텀 확률 분포이며,

는 단어(

)에 배분된 주제 이외의 모든 주제 변수이며,

는 단어(

)에 배분된 저자 이외의 모든 저자 변수이며，

는 텀(v)이 주제(

)에 배분되는 차수이며,

는 저자(

)가 책임진 단어가 주제(k)에 배분되는 차수이다.K is the number of subjects covered by at least two scientific research articles collected, Nm is the number of words in the scientific research document (m), and a _m is a vector formed by the author of the scientific research document (m) A _m is the number of authors of the scientific research literature (m)

Is a temporally varying Beta distribution parameter vector of the subject k,

Is the subject allocation of the n words of the scientific research document (m)

Is the authorship of the n words of the scientific research literature (m)

Is the n-th word of the scientific research document (m)

Is the time stamp of the n words of the scientific research document (m)

The

(A = 1, ... , A) is a Dirichlet priori parameter vector,

Is the subject probability distribution parameter of the author (a)

The

(K = 1, ... , ≪ / RTI > K,

Is the term probability distribution of the subject (k)

Is the word (

) Are all subject variables other than those that are distributed,

Is the word (

) Are all authors other than authors,

(V) is the subject (

), ≪ / RTI >

The author (

) Is the order to which the words responsible are allocated to the topic (k).

Preprocessing the scientific research literature through a preprocessing unit, constructing an authorship evolution model of two or more scientific research documents, and entering step 103.

Step 103: Establish scientific research information estimation parameters through the acquisition unit, and obtain scientific research information evolution results by computing scientific research estimation parameters using evolutionary arithmetic formulas.

), 저자 주제 파라미터(

), 주제 시간 파라미터(

), 과학 연구 문헌(m) 중 제 n개 단어의 주제 배분(

) 및 과학연구 문헌(m) 중 제 n개 단어의 저자 배분(

)을 포함한다. 그 중, 주제 텀 파라미터(

), 저자 주제 파라미터(

), 과학 연구 문헌(m) 중 제 n개 단어의 주제 배분(

) 및 과학연구 문헌(m) 중 제 n개 단어의 저자 배분(

)을 설정하는 근사 추론 알고리즘은 변분 기대치 최대화(Variational Expectation-Maximization), 기대치 전파 및 깁스 샘플링 등을 포함한다. 주제 시간 파라미터( )를 설정하는 근사 추론 알고리즘은 최대우도 추정법, 적률 추정법, 퍼트（Program Evaluation and Review Technique，PERT）추정법, TSP（Two-Sided Power Distribution）추산법 및 분위수 추정법 등을 포함한다.In this step, the pre-processing result at step 102 is used to set the scientific research information estimation parameter through an approximate inference algorithm used by the acquisition unit, of which the scientific research information estimation parameter is the subject term parameter

), Author subject parameter (

), Subject time parameters (

), The theme allocation of the n words in the scientific research literature (m)

) And the authors' distribution of the n words in the scientific research literature (m)

). Among them, the theme term parameter (

), Author subject parameter (

), The theme allocation of the n words in the scientific research literature (m)

) And the authors' distribution of the n words in the scientific research literature (m)

) Include approximate inference algorithms, such as Variational Expectation-Maximization, expectation propagation, and Gibbs sampling. The approximate inference algorithm that sets the subject time parameter () includes the maximum likelihood estimation method, the rate estimation method, the program evaluation and review technique (PERT) estimation method, the two-sided power distribution estimation method and the quantile estimation method.

Next, the scientific research information evolution result is obtained by calculating the above scientific research estimation parameter by using the evolutionary calculation formula, among which the evolutionary calculation formula is as follows,

는 과학연구 문헌 저자(a) 주제(k)의 확률 분포이며,

와

는 주제(k)의 시간적 변화 Beta 분포 파라미터이고,

는 베타 분포의 확률 밀도 함수이다.among them,

(A) the probability distribution of the subject (k)

Wow

Is the temporal change Beta distribution parameter of the subject k,

Is the probability density function of the beta distribution.

In this embodiment, the scientific research information evolution result is obtained by calculating the scientific research estimation parameter by using the evolution calculation formula, and the scientific research evolution result is displayed to the user. For example, The results of scientific research information evolution obtained through the above steps show not only the topics of scientific research literature, but also the research interests of researchers, as well as scientific research topics in predefined fields, research interests of researchers Such as the time evolution of the rules can be found.

In this embodiment, two or more scientific research documents are collected in a predetermined field, preprocessing is carried out on scientific research documents, an authors subject evolution model of two or more scientific research documents is constructed, and a scientific research information estimation parameter The results of the evolution of scientific research information are obtained by posting the research topics of the scientific research literature and the research interests of the researchers, It is also advantageous to grasp the temporal change rules such as the theme of scientific research in the scientific research field and the research interest of the researcher, and to find out the evolutionary fallacy of the scientific research theme in the predetermined field.

FIG. 2 is a flowchart of a second embodiment of a method for analyzing the evolution of scientific research information according to the present invention. FIG. 3 is a probability diagram of the AToT model in the present embodiment. FIG. FIG. 5 is a graph showing the temporal change of the theme 'EM and mixed model' in the present embodiment. FIG. 6 is a graph showing the temporal change of the theme 'visual identification and factor analysis' , FIG. 7 is a trend diagram of the temporal change of the subject 'data model and learning algorithm' in the present embodiment, FIG. 8 is a trend of temporal change of the 'support vector machine and kernel technique' FIG. 10 is a graph showing the temporal change of the theme 'language identification' in the present embodiment, and FIG. 11 is a graph showing the change of the theme 'Bayesian learning' in the present embodiment. FIG. 12 is a graph showing a change in the time-dependent change FIG. 13 is a time evolution diagram of the research subject of the researcher in this embodiment, and FIG. 14 is a schematic diagram of the perplexity of the AT model and the AToT model in this embodiment. As shown in Fig. 2, the operation process of the analysis method of scientific research information evolution of the present embodiment specifically includes the following steps.

Step 201: Collect more than one scientific research literature in a pre-established field through the collection unit.

In this example, a total of 13 articles on the Neural Information Processing System (NIPS) from 1987 to 1999 were collected. The volume of scientific research literature is 1740, Step 202 is entered.

Step 202: The preprocessing unit preprocesses the scientific research literature.

In this step, all the scientific research articles collected through the preprocessing unit are preprocessed, including filtering out of use words, numbers and words lower than a predetermined frequency, and the name of the author of the scientific research document The method of resolving the name of the name includes the artificial neutralization method, the semi-automatic neutralization method, the automatic neutralization method and the mixed neutralization method. And a set of science and technology workers who are made up of A authors. In addition, the presentation time of the scientific research literature is normalized to the interval (0, 1). In the present embodiment, the preprocessing unit preprocesses the collected NIPS papers, obtains a set of words consisting of V terms after completion of preprocessing, and a set of scientific and technological workers composed of authors of A. In addition, The presentation time of the paper is normalized to the interval (0, 1), where V = 12364 and A = 2037. After preprocessing the scientific research literature, step 203 is entered.

Step 203: Build the authors subject evolution model of the collected scientific research literature.

), 저자 주제 파라미터(

), 과학연구 문헌(m) 중 제 n개 단어의 주제 배분(

) 및 과학연구 문헌(m) 중 제 n개 단어의 저자 배분(

)을 계산하는 것을 포함한다. 그 중, 깁스 샘플링 알고리즘 공식은 아래와 같으며, In this step, the AToT model of the collected scientific research literature is constructed. The probability of the AToT model is as shown in FIG. 3. Specifically, the Gibbs sampling algorithm is used to calculate the theme term parameter

), Author subject parameter (

), The theme allocation of the n words in the scientific research literature (m)

) And the authors' distribution of the n words in the scientific research literature (m)

). ≪ / RTI > Among them, the formula of Gibbs sampling algorithm is as follows,

는 주제(k)의 시간적 변화의 베타(Beta) 분포 파라미터 벡터이며，

는 과학연구 문헌(m) 중 제 n개 단어의 주제 배분이고,

는 과학연구 문헌(m) 중 제 n개 단어의 저자 배분이며,

는 과학연구 문헌(m) 중 제 n개 단어이며,

는 과학연구 문헌(m) 중 제 n개 단어의 타임 스탬프이며,

는

（a = 1, …, A）의 디리클레(Dirichlet) 선험 파라미터 벡터이며,

는 저자 (a)의 주제 확률 분포 파라미터이며,

는

(k = 1, …, K）의 디리클레 선험 파라미터 벡터이며,

는 주제(k)의 텀 확률 분포이며,

는 단어(

)에 배분된 주제 이외의 모든 주제 변수이며,

는 단어(

)에 배분된 저자 이외의 모든 저자 변수이며，

는 텀(v)이 주제(

)에 배분되는 차수이며,

는 저자(

)가 책임진 단어가 주제(k)에 배분되는 차수이며, 그 중, K=100,

=50/K（

），

=0.1（

）으로 설정할 수 있다. 두편 이상의 과학연구 문헌의 저자 주제 진화 모형을 구축한 후, 단계 204에 진입한다.K is the number of subjects covered by at least two scientific research articles collected, Nm is the number of words in the scientific research document (m), and a _m is a vector formed by the author of the scientific research document (m) A _m is the number of authors of the scientific research literature (m)

Is a Beta distribution parameter vector of the temporal change of the subject k,

Is the subject allocation of the n words of the scientific research document (m)

Is the authorship of the n words of the scientific research literature (m)

Is the n-th word of the scientific research document (m)

Is the time stamp of the n words of the scientific research document (m)

The

(A = 1, ... , A) is a Dirichlet priori parameter vector,

Is the subject probability distribution parameter of the author (a)

The

(k = 1, ..., K)

Is the term probability distribution of the subject (k)

Is the word (

) Are all subject variables other than those assigned to the subject,

Is the word (

) Are all authors other than authors,

(V) is the subject (

), ≪ / RTI >

The author (

) Is the order to which the word responsible is allocated to the subject (k), among which K = 100,

= 50 / K (

),

= 0.1 (

). After constructing the authorship evolution model of more than one scientific research literature, enter step 204.

Step 204: Set the scientific research information estimation parameter through the acquisition unit.

), 저자 주제 파라미터(

), 주제 시간 파라미터(

), 과학 연구 문헌(m) 중 제 n개 단어의 주제 배분(

) 및 과학연구 문헌(m) 중 제 n개 단어의 저자 배분(

)을 포함한다. In this step, by using the AToT model constructed at step 203, the pre-processing result at step 202 is used to set scientific research information estimation parameters through an approximate inference algorithm used by the acquisition unit, The information estimate parameter is the subject term parameter (

), Author subject parameter (

), Subject time parameters (

), The theme allocation of the n words in the scientific research literature (m)

) And the authors' distribution of the n words in the scientific research literature (m)

).

)에 상응한 주제(

)와 저자(

)를 배분하며, 디리클레 분포의 기대치를 이용하여 주제(k)의 텀 확률분포 파라미터(

) 및 저자(a)의 주제 확률 분포 파라미터(

)를 추정하며, (

) 및 (

)의 계산 공식은 아래와 같으며,In the present embodiment, Gibbs sampling is repeated 2000 times, and the whole process shows a tendency to converge,

) Corresponding topic (

) And author

) And distributes the term probability distribution parameter (k) of the subject (k) using the dirichlet distribution expectation

) And the subject's probability distribution parameter (a)

), And

) And (

) Is as follows,

의 계산 속도를 제고시키기 위하여, 적률 추정법을 이용하여 주제(k)의 시간적 변화 Beta 분포 파라미터 벡터(

)를 계산하며, 계산 공식은 아래와 같으며,

In order to improve the computation speed of the subject (k), the temporal change Beta distribution parameter vector

), And the calculation formula is as follows,

는 미리 설정한 분야의 주제(k)의 샘플링 분산이며, 상기 t_k와

의 계산 공식은 아래와 같으며,Among them, t _k is a sampling average value of the subject (k) of a preset area,

Is the sampling distribution of the subject (k) of a preset field, and the t _k

The calculation formula is as follows,

는 과학연구 문헌(m) 중의 단어가 주제(k)에 배분되는 차수이다.among them,

Is a degree to which the words in the scientific research document (m) are distributed to the subject (k).

In this step, the author's subject evolutionary model is used to calculate 100 subjects from the 1740 scientific research literature. The description of each subject is divided into three parts: (a) all 10 terms and the corresponding percentage; (b) a total of ten researchers and corresponding probabilities that are most relevant to the subject; (c) Includes temporal trends in the subject. Eight of the 1,740 scientific research documents obtained from the calculations are shown in Table 1.

Topic 11
'Enhanced learning' Topic 88
'EM and mixed model' Term percentage Term percentage state 0.0468466 density 0.0279477 learning 0.0252876 log 0.0217790 belief 0.0213999 distribution 0.0186946 policy 0.0182191 mixture 0.0178379 function 0.0175122 method 0.0144108 action 0.0150383 gaussian 0.0142394 states 0.0148615 likelihood 0.0140681 reinforcement 0.0118574 entropy 0.0132113 actions 0.0118574 gaussians 0.0123546 mdp 0.0102670 form 0.0113264 author percentage author percentage Zhang_N 0.629412 Barron_A 0.608507 Rodriguez_A 0.578235 Wainwright_M 0.372871 Dietterich_T 0.342954 Mukherjee_S 0.340927 Sallans_B 0.228042 Li_J 0.337108 Walker_M 0.189143 Jebara_T 0.253203 Koller_D 0.1885150 Millman_K 0.171569 Yeung_D 0.1213730 Fisher_J 0.148230 Thrun_S 0.0842081 Ihler_A 0.128369 Konda_V 0.0680365 Beal_M 0.126578 Parr_R 0.0468006 Hansen_L 0.0849109 Topic 51
'Visual identification and factor analysis' Topic 58
'Data Model and Learning Algorithm' Term percentage Term percentage sejnowski 0.0265409 learning 0.00904655 eye 0.0265409 model 0.00752741 ica 0.0183324 neural 0.00705102 vor 0.0159531 data 0.00700339 disparity 0.0153583 function 0.0068393 head 0.0135738 network 0.00624646 position 0.0125031 input 0.00593946 eeg 0.0119083 set 0.00561128 parietal 0.0109566 networks 0.00556365 salk 0.0105997 figure 0.00545249 author percentage author percentage Sejnowski_T 0.410459 Gray_M 0.974482 Pouget_A 0.269781 Dimitrov_A 0.973538 Anastasio_T 0.112957 Galperin_G 0.97094 Horiuchi_T 0.0328485 Malik_J 0.968536 Albright_T 0.0099278 Davies_S 0.966534 Jousmaki_V 0.00791139 Cook_G 0.96519 Fredholm_H 0.00681818 Ghosn_J 0.964184 Bohr_J 0.00643777 Orponen_P 0.964184 Ramanujam_N 0.00621891 Yen_S 0.963001 Dixon_L 0.00585938 Chatterjee_C 0.962627 Topic 87
'Support vector machine and kernel technique' Topic 37
'Neural network' Term percentage Term percentage set 0.0188195 learning 0.01106740 support 0.0187117 network 0.00948016 vector 0.0186039 neural 0.00780503 kernel 0.0160163 input 0.00682192 function 0.0146146 model 0.00681643 svm 0.0138060 training 0.00604202 training 0.0129974 data 0.00597611 problem 0.0124583 figure 0.00594316 space 0.0119731 networks 0.00560813 solution 0.0115957 function 0.00554222 author percentage author percentage Scholkopf_B 0.949692 Reggia_J 0.979832 Crisp_D 0.888975 Todorov_E 0.976750 Laskov_P 0.706170 Horne_B 0.974146 Steinhage_V 0.634973 Thmn_S 0.973083 Chapelle_O 0.610385 Weigend_A 0.972806 Li_Y 0.513418 McCallum_R 0.969777 Herbrich_R 0.454384 Camana_R 0.969388 Gordon_M 0.425090 Slaney_M 0.969382 Vapnik_V 0.330421 Miikkulainen_R 0.968541 Dom_B 0.286036 Bergen_J 0.968358 Topic 47
"Language Identification" Topic 78
'Bayesian learning' Term percentage Term percentage hmm 0.0415364 bayesian 0.0243032 speech 0.0392921 sampling 0.018456 hmms 0.0216579 prior 0.0178563 mixture 0.0179708 distribution 0.0148578 suffix 0.0104362 monte 0.0127588 probabilistic 0.00995527 carlo 0.0118592 probabilities 0.00947434 model 0.0109597 singer 0.0088331 posterior 0.0105099 acoustic 0.0088331 priors 0.00946041 saul 0.00867279 sample 0.00901063 author percentage author percentage Rigoll_G 0.460882 Schuurmans_D 0.651505 Singer_Y 0.437547 Sykacek_P 0.495506 Nix_D 0.192342 Andrieu_C 0.413324 Saul_L 0.170699 Rasmussen_C 0.344185 Hermansky_H 0.0795602 Zlochin_M 0.244745 Roweis_S 0.0391364 Beal_M 0.157807 Attias_H 0.0357538 Hansen_L 0.122773 Movellan_J 0.033414 Herbrich_R 0.0882701 Schuster_M 0.0293324 Downs_O 0.0694726 Muller_K 0.028258 Williams_C 0.0652069

Sets the scientific research information estimation parameters through the acquisition unit, calculates 100 subject terms probability, author subject probability and subject time parameters of the scientific research literature of 1740, and then enters step 205.

Step 205: Obtain scientific research information evolution results by computing scientific research estimation parameters using the evolutionary arithmetic formula

In this step, the scientific research information evolution result is obtained by calculating the scientific research estimation parameter by using the evolution calculation formula. The evolution calculation formula is as follows,

는 과학연구 문헌의 저자(a) 주제(k)의 확률이며,

와

는 주제(k)의 시간적 변화의 Beta 분포 파라미터이고,

는 Beta 분포의 확률 밀도 함수이다.among them,

(A) the probability of the subject (k) of the scientific research literature,

Wow

Is the Beta distribution parameter of the temporal change of the subject (k)

Is a probability density function of Beta distribution.

)에 관하여 크기가 K x (1999 1987+1)=100 x 13인 매트릭스를 얻으며, 해당 매트릭스의 각 원소는 해당 연구원이 특정 시점에 특정 주제에 대한 관심도이다. 표 1의 연구원 Sejnowski를 예로 들어 기술방안을 설명한다.Each researcher (

), A matrix with a size of K x (1999 1987 + 1) = 100 x 13 is obtained, and each element of the matrix is interested in a particular subject at a particular point in time. Using the example of Sejnowski, the researcher in Table 1 as an example, we describe the technical solution.

Researcher Sejnowski published a total of 43 scientific research papers at the NIPS meeting between 1987 and 1999. Among them, the time distribution diagrams of 43 scientific research documents are shown in Fig. 12, and Sejnowski 13 shows the time evolution model of the research topic. FIG. 13 shows the time evolution of the research topic at different times using Hinton diagram, The larger the area of the rectangle, the greater the interest of Sejnowski in research on the subject. As can be seen from FIG. 13, Sejnowski's research topics from 1987 to 1999 mainly consisted of 'visual recognition and factor analysis' (topic 51), 'neural network' (topic 37) and 'data model and learning algorithm' Sejnowski's research interest in the early years (1989 to 1993) is 'visual recognition and factor analysis', and since 1994, Sejnowski's research topics are 'neural network' (1994) Learning algorithm '(1996), and the research intensity is relatively high (the number of documents published increases). Since 1997, Sejnowski 's research topic has returned to' visual recognition and factor analysis', and the intensity of research has dropped somewhat. During the entire period from 1987 to 1999, Sejnowski conducted research on his main research topic, 'Visual perception and factor analysis'.

In this embodiment, the scientific research information evolution result is obtained by calculating the scientific research estimation parameter by using the evolution calculation formula, and the scientific research evolution result is displayed to the user. For example, The results of scientific research information evolution obtained through the above steps show not only the topic of scientific research literature, but also the research interests of researchers, as well as scientific research topics in pre-established scientific research fields, Research interests and other temporal evolutionary rules.

)에 관하여, 혼란도 계산 공식은 아래와 같으며,In practical applications, perplexity is a measure of the ability to model generalizability, and the smaller the value of confusion, the stronger the ability to generalize the model. In order to evaluate the generalization ability of the evolutionary AToT model, the present embodiment divides 1740 scientific research documents into two parts, 1557 documents as a training set, 183 documents as a test set , The test set contains 102 single author scientific research papers, and all authors from the test set are also derived from the training set. In the AToT model, the scientific research literature in the test set (

), The confusion degree calculation formula is as follows,

among them,

,

및

에 의하여, 혼란도 계산 공식 중의

,

및

의 구체적인 값을 추산할 수 있으며, 본 실시예는 테스트 집합에 대해 깁스 샘플링을 S(예를 들어 S=10)차 실행하며, 혼란도는 S차 샘플링의 평균값을 취한다. 도 14로부터 알 수 있다시피, 주제 갯수가 10을 초과할 때, AToT 모형의 혼란도는 AT 모형보다 현저히 작으며, 이는 AToT 모형의 성능이 AT 모형보다 우수하다는 것을 설명한다.Parameters obtained from the training set

,

And

, The confusion degree calculation formula

,

And

And the present embodiment performs Gibbs sampling S (for example, S = 10) for the test set, and the degree of confusion takes an average value of the S-order sampling. As can be seen from FIG. 14, when the number of subjects exceeds 10, the degree of confusion of the AToT model is significantly smaller than that of the AT model, demonstrating that the performance of the AToT model is superior to that of the AT model.

In this embodiment, two or more scientific research documents are collected in a predetermined field, preprocessing is carried out on scientific research documents, an authors subject evolution model of two or more scientific research documents is constructed, and a scientific research information estimation parameter The results of the evolution of scientific research information are obtained by posting the research topics of the scientific research literature and the research interests of the researchers, It is advantageous to grasp the time-varying rules such as the theme of scientific research in the scientific research field and the research interest of the researcher, and to grasp the evolutionary fallacy of the scientific research theme in the predetermined field.

FIG. 15 is a structural diagram of a first embodiment of an apparatus for analyzing scientific research information evolution according to the present invention. FIG. As shown in FIG. 15, the analysis apparatus for scientific research information evolution of the present embodiment includes a collection unit 701, a preprocessing unit 702, and an acquisition unit 703. Among them, the collecting unit 701 is used for collecting two or more scientific research documents in a predetermined field, the preprocessing unit 702 preprocesses the scientific research documents, and the authoring subject evolution models of two or more scientific research documents And the acquisition unit 703 is used to acquire scientific research information evolution results through setting scientific research information estimation parameters and computing scientific research estimation parameters using an evolutionary operation formula.

), 저자 주제 파라미터(

), 과학연구 문헌(m) 중 제 n개 단어의 주제 배분(

) 및 과학연구 문헌(m) 중 제 n개 단어의 저자 배분(

)을 계산하는 것을 포함하며, 그 중, 깁스 샘플링 알고리즘 공식은 아래와 같으며,Among them, the step of the preprocessing unit 702 constructing the authors subject evolution model of two or more scientific research documents uses the Gibbs sampling algorithm formula to construct the subject term parameter

), Author subject parameter (

), The theme allocation of the n words in the scientific research literature (m)

) And the authors' distribution of the n words in the scientific research literature (m)

), Among which the Gibbs sampling algorithm formula is as follows,

는 주제(k)의 시간적 변화의 베타(Beta) 분포 파라미터 벡터이며，

는 과학연구 문헌(m) 중 제 n개 단어의 주제 배분이고,

는 과학연구 문헌(m) 중 제 n개 단어의 저자 배분이며,

는 과학연구 문헌(m) 중 제 n개 단어이며,

는 과학연구 문헌(m) 중 제 n개 단어의 타임 스탬프이며,

는

(a = 1, …, A）의 디리클레(Dirichlet) 선험 파라미터 벡터이며,

는 저자 (a)의 주제 확률 분포 파라미터이며,

는

（k = 1, …, K）의 디리클레 선험 파라미터 벡터이며,

는 주제(k)의 텀 확률 분포이며,

는 단어(

)에 배분된 주제 이외의 모든 주제 변수이며,

는 단어(

)에 배분된 저자 이외의 모든 저자 변수이며，

는 텀(v)이 주제(

)에 배분되는 차수이며,

는 저자(

)가 책임진 단어가 주제(k)에 배분되는 차수이다.K is the number of subjects covered by at least two scientific research articles collected, Nm is the number of words in the scientific research document (m), and a _m is a vector formed by the author of the scientific research document (m) A _m is the number of authors of the scientific research literature (m)

Is a Beta distribution parameter vector of the temporal change of the subject k,

Is the subject allocation of the n words of the scientific research document (m)

Is the authorship of the n words of the scientific research literature (m)

Is the n-th word of the scientific research document (m)

Is the time stamp of the n words of the scientific research document (m)

The

(a = 1, ..., A) is a Dirichlet priori parametric vector,

Is the subject probability distribution parameter of the author (a)

The

(K = 1, ... , ≪ / RTI > K,

Is the term probability distribution of the subject (k)

Is the word (

) Are all subject variables other than those that are distributed,

Is the word (

) Are all authors other than authors,

(V) is the subject (

), ≪ / RTI >

The author (

) Is the order to which the words responsible are allocated to the topic (k).

) 및 저자(a)의 주제 확률 분포 파라미터(

)를 추정하는 것을 포함하며,The step of the acquisition unit 703 setting the scientific research information estimation parameter may be performed using the term probability distribution parameter (k) of the subject (k), specifically using the dirichlet distribution expectation

) And the subject's probability distribution parameter (a)

), ≪ / RTI >

)를 계산하며, 계산 공식은 아래와 같으며,The temporal change of the subject (k) using the ratiometric estimation method. Beta distribution parameter vector (

), And the calculation formula is as follows,

는 미리 설정한 분야의 주제(k)의 샘플링 분산이며, 상기 t_k와

의 계산 공식은 아래와 같으며,Among them, t _k is a sampling average value of the subject (k) of a preset area,

Is the sampling distribution of the subject (k) of a preset field, and the t _k

The calculation formula is as follows,

는 과학연구 문헌(m) 중의 단어가 주제(k)에 배분되는 차수이다.among them,

Is a degree to which the words in the scientific research document (m) are distributed to the subject (k).

In obtaining the scientific research information evolution result by the acquisition unit 703 by calculating the scientific research estimation parameter by using the evolution calculation formula, the evolution calculation formula is as follows,

는 저자(a) 주제(k)의 확률 분포이며,

와

는 주제(k)의 시간적 변화의 Beta 분포 파라미터이다.among them,

Is the probability distribution of the author (a) subject (k)

Wow

Is the Beta distribution parameter of the temporal change of the subject (k).

In this embodiment, two or more scientific research documents are collected in a predetermined field through a collection unit, preprocessing is carried out on a scientific research document through a preprocessing unit, and an author-theme evolution model of two or more scientific research documents is constructed The results of the evolution of scientific research information are obtained through the computation of scientific research estimation parameters by setting the scientific research information estimation parameters through the acquisition unit and using the evolutionary operation formula. , Research interests of researchers can be posted, and time-based change rules such as pre-set scientific research topics in the field of scientific research and research interests of researchers can be discovered and published, and the evolutionary eruption of scientific research topics within a predetermined field can be grasped .

16 is a structural diagram of a second embodiment of an apparatus for analyzing scientific research information evolution according to the present invention. 16, the apparatus for analyzing scientific research information evolution of the present embodiment further includes a display unit 704, the display unit 704 shows the scientific research information evolution result to the user, and the display unit 704 ) Includes a line graph and / or a Hinton diagram, and it is convenient for the user to intuitively check the temporal change rule such as the research topic of the researcher.

Those skilled in the art will appreciate that the present invention may relate to equipment that performs one or more of the operations described herein. The facility may include known facilities among computers that are professionally designed or used universally for the required purpose, and the programs stored in the universally used computer are selectively activated or reconfigured. Such a computer program may be stored on any type of medium that is suitable for storing electronic instructions and is coupled to a bus, and which may be stored on a readable medium of a facility (e.g., a computer) (RAM), read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), and the like, as well as any type of disk (including floppy disk, hard disk, laser disk, CD-ROM and magneto-optical disk) ROM), Electrically Erasable Programmable ROM (EEPROM), a flash memory, a magnetic card or an optical card. Readable media includes any structure for storing or transmitting information in a facility (e.g., computer) readable format. For example, the readable medium can be a random access memory (RAM), a read only memory (ROM), a magnetic disk storage medium, an optical storage medium, a flash storage device, a signal propagated in electrical, optical, A carrier wave, an infrared signal, a digital signal).

It will be appreciated by those skilled in the art that the computer program instructions may be embodied in block diagrams and / or block diagrams and / or block diagrams, and combinations of blocks in the block diagrams and / or block diagrams and / or flowcharts . Such computer program instructions may be provided to a processing device of a computer, a dedicated computer, or other programmable data processing method that is commonly used to create a device, and thereby to execute a command through a processing device of a computer or other programmable data processing method Block diagrams and / or flowchart diagrams, and methods specified in blocks or blocks.

It will be apparent to those skilled in the art that various operations, methods, steps in the process, measures, and methods discussed in the present invention may be altered, altered, combined, or eliminated. Further, various operations, methods, and other steps, measures, and measures in the processes already discussed in the present invention may be replaced, changed, rearranged, disassembled, combined, or deleted. Further, steps, measures, and measures in various operations, methods, and processes disclosed in the present invention may be replaced, changed, rearranged, disassembled, combined, or deleted.

It will be apparent to those skilled in the art that various modifications and variations can be made without departing from the principles of the invention, do.

Claims (11)

In an analysis method of scientific research information evolution, the analysis method of scientific research information evolution is:
Collecting at least two scientific research documents in a predetermined field through a collection unit;
Preprocessing the scientific research literature through a preprocessing unit and building an authorship evolution model of the two or more scientific research literature; And
Acquiring a scientific research information evolution result by setting a scientific research information estimation parameter through an acquisition unit and calculating the scientific research information estimation parameter using an evolutionary calculation formula,
The step of constructing the authors' topic evolution model of the above two or more scientific research documents is carried out by using the Gibbs sampling algorithm formula,

), Author subject parameter (

), The theme allocation of the n words in the scientific research literature (m)

) And the authors' distribution of the n words in the scientific research literature (m)

), ≪ / RTI >
The formula for the Gibbs sampling algorithm is as follows,

K is the number of subjects covered by at least two scientific research articles collected, N _m is the number of words in the scientific research document (m), and a _m is the number of scientific research articles (m) A _m is the number of author of the scientific research document (m)

Is a temporally varying Beta distribution parameter vector of the subject k,

Is the subject allocation of the n words of the scientific research document (m)

Is the authorship of the n words of the scientific research literature (m)

Is the n-th word of the scientific research document (m)

Is the time stamp of the n words of the scientific research document (m)

The

(A = 1, ... , A) is a Dirichlet priori parameter vector,

Is the subject probability distribution parameter of the author (a)

The

(K = 1, ... , ≪ / RTI > K,

Is the term probability distribution of the subject (k)

Is the word (

) Are all subject variables other than those that are distributed,

Is the word (

) Are all authors other than authors,

(V) is the subject (

), ≪ / RTI >

The author (

) Is a degree distributed to the subject (k)
The step of setting the scientific research information estimating parameter is specifically a method of estimating the term probability distribution parameter (k) of the subject (k) using the dirichlet distribution expectation

) And the subject's probability distribution parameter (a)

), And the calculation formula is as follows,

The temporal change of the subject (k) using the ratiometric estimation method. Beta distribution parameter vector (

), And the calculation formula is as follows,

Among them, t _k is a sampling average value of the subject (k) of a preset area,

Is the sampling distribution of the subject (k) of a preset field, and the t _k

The calculation formula is as follows,

among them,

Characterized in that words in the scientific research document (m) are orders distributed to the subject (k). The method according to claim 1,
Further comprising displaying the scientific research information evolution result to a user through a display unit, wherein the display scheme is a line graph and / or a Hinton diagram. The method according to claim 1,
The step of preprocessing the scientific research document through the preprocessing unit may be,
Filtering words that are less than a predetermined frequency and a word that is not in use in the scientific research literature and proceeding to solve the hypothesis for the author name of the scientific research literature; And
Obtaining the V terms of the A authors of the two or more scientific research documents and performing a normalization process on the presentation times of the two or more scientific research documents; Analysis method. The method according to claim 1,
Wherein the step of obtaining the evolution result of the scientific research information by computing the scientific research information estimation parameter using the evolution computation formula is as follows,

among them,

Is the probability of the author (a) subject (k)

Wow

Is a temporal change Beta distribution parameter of the subject (k). An apparatus for analyzing the evolution of scientific research information, comprising:
A collection unit, a preprocessing unit and an acquisition unit,
The collection unit collecting two or more scientific research documents in a predetermined field;
Wherein the preprocessing unit preprocesses the scientific research literature and builds an authorship evolution model of the two or more scientific research literature;
Wherein the acquisition unit sets the scientific research information estimation parameter and acquires the scientific research information evolution result by calculating the scientific research information estimation parameter using the evolutionary operation formula,
The step of the preprocessing unit constructing an authorship evolution model of the two or more scientific research documents uses a Gibbs sampling algorithm formulation to generate a topic term parameter

), Author subject parameter (

), The theme allocation of the n words in the scientific research literature (m)

) And the authors' distribution of the n words in the scientific research literature (m)

), ≪ / RTI >
The formula for the Gibbs sampling algorithm is as follows,

K is the number of subjects covered by at least two scientific research articles collected, Nm is the number of words in the scientific research document (m), and a _m is a vector formed by the author of the scientific research document (m) A _m is the number of authors of the scientific research literature (m)

Is a temporally varying Beta distribution parameter vector of the subject k,

Is the subject allocation of the n words of the scientific research document (m)

Is the authorship of the n words of the scientific research literature (m)

Is the n-th word of the scientific research document (m)

Is the time stamp of the n words of the scientific research document (m)

The

(A = 1, ... , A) is a Dirichlet priori parameter vector,

Is the subject probability distribution parameter of the author (a)

The

(K = 1, ... , ≪ / RTI > K,

Is the term probability distribution of the subject (k)

Is the word (

) Are all subject variables other than those that are distributed,

Is the word (

) Are all authors other than authors,

(V) is the subject (

), ≪ / RTI >

The author (

) Is a degree distributed to the subject (k)
The step of the acquiring unit setting the scientific research information estimation parameter may be performed by using the term probability distribution parameter (k) of the subject (k)

) And the subject's probability distribution parameter (a)

), And the calculation formula is as follows,

The temporal change of the subject (k) using the ratiometric estimation method. Beta distribution parameter vector (

), And the calculation formula is as follows,

Among them, t _k is a sampling average value of the subject (k) of a preset area,

Is the sampling distribution of the subject (k) of a preset field, and the t _k

The calculation formula is as follows,

among them,

Is an order in which the words in the scientific research document (m) are distributed to the subject (k). The method of claim 5,
Wherein the scientific research information evolution result is displayed to the user, and the display method further includes a line graph and / or a Hinton diagram display unit. The method of claim 5,
Wherein the obtaining unit obtains a scientific research information evolution result by calculating the scientific research information estimation parameter using an evolution computation formula, the evolution computation formula is as follows,

among them,

Is the probability of the author (a) subject (k)

and

Is a temporal change Beta distribution parameter of the subject (k). delete delete delete delete

Images