CN110046713A - Robustness sequence learning method and its application based on multi-objective particle swarm optimization - Google Patents

Abstract

The present invention relates to a kind of robustness sequence learning method and its application based on multi-objective particle swarm optimization, the following steps are included: step 1, based on deviation-variance isostatic theory, design the validity departure function and robustness variance function of order models, two optimality criterions of building sequence study；Step 2 is based on multi-objective particle frame on sequence learning data set, and the validity departure functions of iteration optimization order models and robustness variance function the two targets are to train order models, to generate order models filing disaggregation；Step 3, based on the thought of the preference ranking organization methods for enrichment evaluations PROMETHEE II in Multiple Attribute Decision Making Theory, the order models filing solution caused by the previous step concentrates selection one order models for having the Pareto of maximum " net flow " ranking value optimal in this, as the final order models trained.Compared with prior art, the present invention has many advantages, such as to improve whole user satisfaction, enhances user experience.

Description

Robustness sequencing learning method based on multi-objective particle swarm optimization and application thereof

Technical Field

The invention relates to the field of information retrieval and machine learning, in particular to a robustness sequencing learning method based on multi-objective particle swarm optimization and application thereof.

Background

The sequencing learning is to use the machine learning technology to automatically train out a sequencing model to solve the sequencing problem. The method is a hotspot problem researched in the field of information retrieval and machine learning, and has wide application prospects in the aspects of information retrieval, search engines, recommendation systems, question-answering systems and the like.

Due to the dynamics of the Web and the diversity of the user information requirements, the performance of some Web search queries may vary greatly and may suffer from significant loss under different ranking models, thereby reducing the user experience. A robust retrieval system should ensure that the user experience is not compromised by the presence of poorly performing queries. Therefore, in order to improve the overall user experience as much as possible, in addition to the traditional relevance and importance criteria, how to ensure the robustness of the ranking model, i.e. relative to a simple benchmark, although an average gain is obtained overall, the new ranking model usually suffers from the performance loss of many queries, which is an important issue faced by the ranking learning research in recent years. Therefore, it is essential to develop a robust ranking learning method to train a robust perceptual ranking model to improve the overall satisfaction of all users as much as possible.

Currently, rank learning researchers compare and evaluate multiple ranking models, mainly by designing advanced ranking features and/or by developing advanced rank learning methods, selecting a best-available ranking model based on some effectiveness metrics, such as Normalized Discounted Cumulative Gain (NDCG) and Expected Reciprocal Ranking (ERR), with the goal of focusing on improving the average effectiveness of the ranking model, which often neglects the robustness of the ranking model. The ranking model with poor robustness can cause instability of a ranking system, namely, the performance of some queries is good, while the performance of other queries is poor, so that the ranking result presented to a user is unstable, and the information requirements of different users are difficult to meet as much as possible, thereby being difficult to bring satisfactory experience to the user. Therefore, in the training process of the ranking model, it is necessary to establish an optimization target meeting the actual requirement, and the effectiveness and the robustness of the ranking model are considered to be optimized simultaneously.

Disclosure of Invention

The invention aims to overcome the defects in the prior art and provides a robustness ranking learning method based on multi-objective particle swarm optimization and application thereof.

The purpose of the invention can be realized by the following technical scheme:

a robustness sequencing learning method based on multi-objective particle swarm optimization comprises the following steps:

designing an effective deviation function and a robust variance function of a ranking model based on a deviation-variance balance theory, and constructing two optimization performance indexes of ranking learning;

secondly, training a ranking model by using two targets of an effectiveness deviation function and a robustness variance function of an iterative optimization ranking model on a ranking learning data set based on a multi-target particle swarm optimization algorithm framework, so as to generate a ranking model filing solution set;

and thirdly, based on the idea of a preference order structure evaluation method PROMETHEE II in the multi-attribute decision theory, selecting a Pareto optimal sorting model with the maximum 'net flow' sorting value from the sorting model filing solution set generated in the last step to serve as a trained final sorting model.

Preferably, the two optimization performance indexes for constructing the ranking learning are specifically:

the effectiveness deviation function and the robustness variance function of the query and the query set under the ranking model are respectively defined as follows:

definitions 1 query q_iOf the validity deviation function Bias^R(q_i) Is defined as:

wherein ,representing a query q_iAll documents D under_iThe best effectiveness achieved under the ideal ranking model I, i.e. all documents correctly ranked,representing a query q_iAll documents D under_iActual effectiveness, Bias, obtained under the ranking model R^R(q_i) Representing a query q_iDeviation of actual effectiveness under the ranking model R from the optimal effectiveness of the ideal ranking model I;

definition 2. validity deviation function Bias of query set Q^R(Q) is defined as:

among them, Bias^R(Q) represents all queries Q in a set Q under a ranking model R_iIs the average of the validity deviations, | Q | represents the query Q in the query set Q_iThe total number of (2);

definitions 3 query q_iRobust Variance function Variance of (1)^R(q_i) Is defined as:

Variance^R(q_i)＝[Bias^R(q_i)-Bias^R(Q)]²…(3)

wherein, Variance^R(q_i) Representing queries q under a ranking model R_iIs of validity Bias^R(q_i) Validity deviation Bias from query set Q^RA degree of dispersion of (Q);

definition 4. robust Variance function Variance of query set Q^R(Q) is defined as:

wherein, Variance^R(Q) represents all queries Q in a set Q under a ranking model R_iThe mean of the robust variance of (a);

converting the robustness ranking learning problem into a multi-objective optimization problem considering both effectiveness and robustness, and formally describing the robustness ranking learning problem as follows according to the definition of the effectiveness deviation function and the robustness variance function:

Utility(Q)＝{min Bias^R(Q),min Variance^R(Q)}…(5)

namely, in the process of sequencing learning, the effectiveness deviation function Bias is minimized at the same time^R(Q) and the robust Variance function Variance^R(Q) to train a ranking model, for which purpose the optimization performance index Bias based on the above-constructed ranking model^R(Q) and Variance^R(Q) a multi-objective intelligent optimization algorithm, such as a multi-objective particle swarm optimization algorithm, can be used while minimizing the Bias^R(Q) and Variance^RThe value of (Q) is used for achieving the purpose of balancing the effectiveness and the robustness of the optimized sequencing model.

Preferably, in the multi-objective particle swarm optimization algorithm-based framework, the two objectives of the effectiveness deviation function and the robustness variance function of the iterative optimization ranking model are used to train the ranking model, so that the generation of the archival solution set of the ranking model specifically comprises:

step 1, initializing relevant parameters of a particle swarm;

step 2, based on the given sequencing learning data set, under the ideal sequencing model I, calculating each query q_iBest effectiveness of

Step 3, initializing the relevant information of each particle to generate an initial sequencing model set P;

step 4, initializing an iteration counter t to be 0;

step 5, establishing an initial sequencing model filing solution set Archive, selecting a non-dominant sequencing model from the initial sequencing model P, and storing the non-dominant sequencing model in the sequencing model filing solution set Archive;

step 6, calculating the congestion distance of each non-dominant solution in the archiving solution set Archive of the sequencing model;

step 7, arranging the non-dominant solutions in Archive in a descending order according to the magnitude of the congestion distance;

step 8, performing operation on each particle to update the position and speed information of the particle;

step 9, updating the archiving solution set Archive of the sequencing model;

step 10, updating individual extreme values Pbest [ i ] of each particle in the particle swarm P;

and 11, if t is t +1, if t is less than MAXT, turning to the step 6, otherwise, outputting each sequencing model in the sequencing model Archive solution set Archive, namely generating a final sequencing model set.

Preferably, the relevant parameters of step 1 include population size Pop, acceleration factors c1 and c2, and initial inertia weight ω₀Final inertial weight ω₁Maximum iteration times MAXT, the number N of objective functions and the variation probability Mu.

Preferably, the initializing the relevant information of each particle in step 3 to generate the initial ordering model set P specifically includes:

31) randomly initializing the position P [ i ] of each particle;

real number coding is adopted, and in a feasible ordering model domain of an ordering learning problem, an initial position P [ i ] of each particle is randomly generated, namely a weight corresponding to each ordering characteristic, wherein i is more than or equal to 1 and is less than or equal to Pop;

32) initializing the speed V [ i ] of each particle as 0;

33) calculating an effectiveness deviation function and a robustness variance function value of the sequencing model;

according to the position P [ i ] of each particle]And a linear ranking scoring functionComputing queries q_iEach document d of_ijWherein f is_ijm(q_i,d_ij) Representing query-document pairs (q)_i,d_ij) According to different Score (q)_i,d_ij) Value-from-big to-little for each query q_iLower documents d_ijCarrying out top-n rapid sequencing, and marking Y according to the sequencing position and the relevance of the document_iCombining the ideal ordering model I, respectively calculating the query q according to the formula (1) to the formula (4)_iAnd the validity deviation function Bias of the query set Q^R(q_i) and Bias^R(Q) and a robust Variance function Variance^R(q_i) And Variance^R(Q) to obtain target values for an effectiveness bias function and a robustness variance function of the ranking model;

34) initializing an individual extreme value Pbest [ i ] ═ P [ i ] of the particle;

35) and determining the global extreme value Gbest of the initial population according to the Pbest [ i ] of each particle.

Preferably, the step 8 of performing an operation on each particle to update the position and velocity information of the particle is specifically as follows:

81) randomly selecting a certain particle i from a non-inferior solution set at the front end with larger crowding distance in the ordered sequencing model filing solution set Archive, and setting the position of the certain particle i as a global extremum Gbest;

82) updating the velocity V [ i ] of particle i according to equation (6):

V_im(t+1)＝ω_t*V_im(t)+c1*rand()*[X_im(t)-P_im(t)]+c2*rand()*[X_Gm(t)-P_im(t)]…(6)

wherein the inertia weight at the t-th iterationc1 and c2 are acceleration constantsNumber factor, rand () is [0, 1 ]]Random number of cells, X_im(t) and X_Gm(t) respectively representing individual extreme values Pbest [ i ] of the particle i at the t-th iteration]The m-th dimension component of Gbest and the m-th dimension component of the global extreme value Gbest, wherein the integer i is the particle number and takes the values of 1,2, … … and Pop;

83) updating the position P [ i ] of the particle i according to equation (7):

P_im(t+1)＝P_im(t)+V_im(t+1)…(7)

84) checking whether P [ i ] is within the limits given by the variables, and if the position range of P [ i ] is exceeded, setting the corresponding dimensional variable in P [ i ] as the corresponding boundary value and setting the speed as the reverse direction, namely-V [ i ];

85) if t < MAXT Mu, then executing variation operation on the position P [ i ] of the particle by taking Mu as variation rate;

86) calculating an objective function value of the particle;

according to the position P [ i ] of the particle]And a linear ranking scoring functionComputing queries q_iEach document d of_ijWherein f is_ijm(q_i,d_ij) Representing query-document pairs (q)_i,d_ij) According to different Score (q)_i,d_ij) Value-from-big to-little for each query q_iLower documents d_ijCarrying out top-n rapid sequencing, and marking Y according to the sequencing position and the relevance of the document_iCombining the ideal ordering model I, respectively calculating the query q according to the formula (1) to the formula (4)_iAnd the validity deviation function Bias of the query set Q^R(q_i) and Bias^R(Q) and a robust Variance function Variance^R(q_i) And Variance^R(Q) to obtain target values for a validity deviation function and a robustness variance function of the ranking model.

Preferably, the updating of the Archive solution set Archive of the ranking model in step 9 specifically includes:

if the particles in the population P are not dominated by any particle in the order model Archive solution set Archive, inserting all new non-dominated particles in the order model Archive solution set Archive and deleting all particles dominated by the new particles in the order model Archive solution set Archive, if the order model Archive solution set Archive is full, replacing the particles by the following steps:

step 1, calculating the congestion distance of each non-dominant solution in an Archive solution set Archive of a sequencing model, and arranging the congestion distances in a descending order according to the size of the congestion distance;

and 2, randomly selecting one particle in the non-dominant solution at the front end with smaller crowding distance from the bottom end of the sequencing model Archive solution set Archive, and replacing the particle with a new particle.

Preferably, the step 10 of updating the individual extremum Pbest [ i ] of each particle in the particle group P is specifically:

and comparing the new position of the particle P [ i ] with the advantages and disadvantages of the Pbest [ i ] according to the dominance relation, and updating the individual optimum when the P [ i ] is dominant, namely Pbest [ i ] ═ P [ i ].

Preferably, the third step selects, based on the concept of a preference order structure assessment method prometeei in the multi-attribute decision theory, one Pareto optimal ranking model with the largest "net flow" ranking value from the ranking model filing solution set generated in the previous step, and the selected ranking model is specifically:

step 1, calculating an outflow function of each sequencing model;

respectively calculating the 'outflow' function Out (R) of each sequencing model for the sequencing model filing solution set Archive finally generated in the step two_i) Value, will Out (R)_i) Is defined as:

wherein, | A | represents the total number of Pareto optimized solutions in the ranking model Archive solution set Archive;

step 2, calculating an inflow function of each sequencing model;

respectively calculating the 'inflow' function In (R) of each sequencing model for the sequencing model filing solution set Archive finally generated In the step two_i) Value In (R)_i) Is defined as:

step 3, calculating a net flow function of each sequencing model;

respectively calculating the Net flow function Net (R) of each sequencing model for the sequencing model filing solution set Archive finally generated in the step two_i) Value, will Net (R)_i) Is defined as:

Net(R_i)＝Out(R_i)-In(R_i)

step 4. according to the Net flow function Net (R)_i) Obtaining each sequencing model R in the sequencing model filing solution set Archive_iThe "net flow" ranking value of (a) and a Pareto optimization solution with the largest "net flow" ranking value is selected as the final ranking model R.

The application of the robustness ranking learning method based on multi-objective particle swarm optimization is characterized in that the robustness ranking learning method based on multi-objective particle swarm optimization is applied to a search engine, wherein the search engine comprises Baidu Baigougle, Google must Bing, Sogou and Yahoo, a ranking model trained by the method is embedded into a ranking system of the search engine, and a webpage ranking result of query words required to be searched by a user is predicted by the ranking model, so that the satisfaction degree of the whole user can be improved, the experience of the user is enhanced, and the specific application process is as follows:

step 1, integrating a robustness sequencing learning method based on multi-objective particle swarm optimization into a search engine;

firstly, performing data preprocessing on partial webpages in a search engine webpage index database, and extracting and labeling ranking features of the webpages to construct a search engine ranking learning data set;

secondly, iteratively training a ranking model to generate a robustness-perceived ranking model by using a robustness ranking learning method based on multi-objective particle swarm optimization on the constructed ranking learning data set;

finally, embedding the generated robustness sensing sequencing model into a sequencing system of a search engine;

step 2, executing web page search and presenting a sequencing result;

in a search engine integrated with a robustness ranking learning method based on multi-objective particle swarm optimization, a user can circularly execute webpage search for multiple times;

firstly, a user inputs a query word to be searched in a search box of a search engine and clicks for searching;

secondly, a search engine webpage index database is called by a search engine ranking system, all webpages containing the query word are found out from the database, and the ranking results of webpage search are predicted by calculating which webpages should be ranked in front and which should be ranked behind;

and finally, returning the webpage search sorting result to a search page according to a set mode for presenting to a search user.

Compared with the prior art, the invention has the following advantages:

1) the multi-objective optimization model for sequencing learning is novel.

A multi-target sequencing learning model considering effectiveness and robustness at the same time is built, a deviation-variance equilibrium theory is introduced into a sequencing learning task, a corresponding effectiveness deviation function and a robustness variance function are given to measure the effectiveness and the robustness of the sequencing model respectively, the effectiveness deviation function and the robustness variance function are designed independently, and the weight assignment of a plurality of targets is not considered.

2) The method for sequencing learning is new.

Based on a multi-objective particle swarm optimization framework and the concept of a preference order structure assessment method PROMETHEE II in a multi-attribute decision theory, a robustness ranking learning method based on multi-objective particle swarm optimization is designed to optimize the effectiveness and robustness of a ranking model at the same time, so that the trained ranking model has better balance between effectiveness and robustness, and the overall user experience is improved.

Drawings

FIG. 1 is a flow chart of a robustness ranking learning method based on multi-objective particle swarm optimization according to the present invention;

FIG. 2 is a flow chart of the construction of the order model optimization performance indicators of the present invention;

FIG. 3 is a flowchart illustrating an embodiment of training a ranking model according to the present invention;

FIG. 4 is a flow diagram of a preferred embodiment of the ranking model of the present invention;

FIG. 5 is a schematic diagram of a first embodiment of the present invention;

fig. 6 is a schematic diagram of a second embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, shall fall within the scope of protection of the present invention.

The method aims at the problem of robust sequencing learning, considers the effectiveness and robustness of a sequencing model at the same time, optimizes the sequencing model respectively, models the robust sequencing learning problem into a multi-objective optimization problem, gives a corresponding effectiveness deviation function and a robustness variance function to measure the effectiveness and robustness of the sequencing model respectively based on a deviation-variance equilibrium theory, and designs a robust sequencing learning method based on multi-objective particle swarm optimization algorithm framework and a PROMETHEE II idea of a preference order structure evaluation method in a multi-attribute decision theory.

The existing sequencing learning method mainly models the sequencing learning problem into the problems of classification, regression, ordinal regression, convex optimization and the like, and the adopted technologies mainly comprise a support vector machine, a neural network, a multiple addition regression tree, an extreme learning machine, Boosting, genetic programming and the like, and the technologies are difficult to simultaneously optimize a plurality of performance indexes. Compared with the existing sequencing learning method, the robust sequencing learning method based on the multi-target particle swarm optimization is designed aiming at the robust sequencing learning. The sequencing learning method models a sequencing learning problem into a multi-objective optimization problem, and the adopted technology is a multi-objective particle swarm optimization technology. The method designs a plurality of objective functions independently without assigning weights of a plurality of objectives, and can simultaneously optimize the effectiveness and robustness of the ranking model in the training process of the ranking model, so that the trained ranking model has better balance between the effectiveness and the robustness, thereby improving the satisfaction experience of the whole user.

The robustness ranking learning method based on multi-objective particle swarm optimization simultaneously considers the effectiveness and robustness of a ranking model and constructs a ranking learning task as a multi-objective optimization problem; based on a deviation-variance equilibrium theory, designing a corresponding effectiveness deviation function and a robustness variance function to respectively measure the effectiveness and the robustness of the ranking model, wherein the effectiveness deviation function is used for measuring the effectiveness of the ranking model, and the robustness variance function is used for measuring the robustness of the ranking model; designed based on the multi-objective particle swarm optimization algorithm framework and the PROMETHEE II idea of the preference order structure evaluation method in the multi-attribute decision theory.

The flow framework of the robustness ranking learning method based on multi-objective particle swarm optimization is shown in FIG. 1, and mainly comprises three stages of construction of optimization performance indexes of a ranking model, training of the ranking model and optimization of the ranking model. Firstly, designing an effective deviation function and a robust variance function of a ranking model based on a deviation-variance balance theory, and constructing two optimization performance indexes of ranking learning; secondly, on the basis of a multi-objective particle swarm optimization algorithm framework, training a ranking model by iteratively optimizing two targets, namely an effectiveness deviation function and a robustness variance function of the ranking model on the ranking learning data set, so as to generate a ranking model filing solution set; and finally, based on the idea of a preference order structure assessment method PROMETHEE II in the multi-attribute decision theory, selecting a Pareto optimal sorting model with the maximum 'net flow' sorting value from the sorting model filing solution set generated in the last stage to serve as a trained final sorting model.

The robustness ranking learning method based on multi-target particle swarm optimization mainly comprises three stages of construction of the optimization performance index of a ranking model, training of the ranking model and optimization of the ranking model, wherein the specific steps of each stage are as follows:

stage one: and constructing an optimized performance index of the sequencing model.

Based on a deviation-variance equilibrium theory, designing an effectiveness deviation function and a robustness variance function of a ranking model to construct two optimization performance indexes of ranking learning: and the effectiveness and the robustness of the sequencing model are represented.

The steps of constructing the optimized performance index of the ranking model are shown in fig. 2, and the effectiveness deviation function and the robustness variance function of the query and the query set under the ranking model are respectively defined in each step and are set forth as follows:

step 1, constructing an effectiveness deviation function of query.

Definitions 1 query q_iOf the validity deviation function Bias^R(q_i) Is defined as:

wherein ,representing a query q_iAll documents D under_iThe best effectiveness achieved under the ideal ranking model I, i.e. all documents correctly ranked,representing a query q_iAll documents D under_iActual effectiveness, Bias, obtained under the ranking model R^R(q_i) Representing a query q_iDeviation of actual effectiveness under ranking model R from the optimal effectiveness of ideal ranking model I. Here, the validity may refer to some commonly used performance indicators in the field of information retrieval, such as Normalized Discounted Cumulative Gain (NDCG), Expected Reciprocal Rank (ERR), and the like, and the calculation formula of each validity indicator is omitted here.

And 2, constructing an effectiveness deviation function of the query set.

Definition 2. validity deviation function Bias of query set Q^R(Q) is defined as:

among them, Bias^R(Q) represents all queries Q in a set Q under a ranking model R_iIs the average of the validity deviations, | Q | represents the query Q in the query set Q_iThe total number of (c).

And 3, constructing a robustness variance function of the query.

Definitions 3 query q_iRobust Variance function Variance of (1)^R(q_i) Is defined as:

Variance^R(q_i)＝[Bias^R(q_i)-Bias^R(Q)]²…(3)

wherein, Variance^R(q_i) Representing queries q under a ranking model R_iIs of validity Bias^R(q_i) Validity deviation Bias from query set Q^RDegree of dispersion of (Q).

And 4, constructing a robust variance function of the query set.

Definition 4. robust Variance function Variance of query set Q^R(Q) is defined as:

wherein, Variance^R(Q) represents all queries Q in a set Q under a ranking model R_iIs the average of the robust variance of (a).

Converting the robustness ranking learning problem into a multi-objective optimization problem considering both effectiveness and robustness, and formally describing the robustness ranking learning problem as follows according to the definition of the effectiveness deviation function and the robustness variance function:

Utility(Q)＝{min Bias^R(Q),min Variance^R(Q)}…(5)

namely, in the process of sequencing learning, the effectiveness deviation function Bias is minimized at the same time^R(Q) and the robust Variance function Variance^R(Q) to train the ranking model. Therefore, the optimization performance index Bias based on the sequencing model constructed in the above way^R(Q) and Variance^R(Q) a multi-objective intelligent optimization algorithm, such as a multi-objective particle swarm optimization algorithm, can be used to simultaneously minimize the Bias^R(Q) and Variance^RThe value of (Q) is used for achieving the purpose of balancing the effectiveness and the robustness of the optimized sequencing model.

And a second stage: and training a sequencing model.

Based on a multi-objective particle swarm optimization algorithm framework, two objective functions of the effectiveness deviation and the robustness variance of the ordering model designed in the previous stage are optimized on the ordering learning data set to iteratively train the ordering model, so that an archiving solution set of the ordering model is generated.

In a specific implementation, for the ordering learning problem, real number encoding is adopted to represent particles, one particle represents an ordering model, and the search space of the particles is M-dimensional (i.e., each particle is composed of M-dimensions, and M represents the total dimension of the ordering features to be considered in the ordering model). The position P and the speed V of the particle can be represented by an M-dimensional vector, the position P of the particle represents the weight corresponding to each sorting feature in the sorting model, and the position P [ i ] of the particle i]Is composed of the weights corresponding to the M-dimensional sorting features in the ith sorting model, i.e. using the vector P [ i]＝(P_i1,P_i2,P_i3,…,P_im,…,P_iM)^TIs shown in which P is_imRepresenting the weight corresponding to the M-dimension sorting feature in the ith sorting model, wherein the integer i is 1,2, … and Pop, the integer M is 1,2, … and M, the integer Pop represents the total size of the particle population, and the integer M represents the total dimension of the sorting feature in the sorting model; the velocity V of the particle is used to change the position P of the particle, i.e. to adjust the weight of each ordering feature in the ordering model, and similarly to the definition of the position of the particle, the velocity V [ i ] of the particle i]Using the vector V [ i ]]＝(v_i1,v_i2,v_i3,…,v_im,…,v_iM)^TTo indicate.

In the robustness ranking learning method based on multi-objective particle swarm optimization, the specific implementation flow of training of the ranking model is shown in fig. 3, and the detailed steps are as follows:

step 1, initializing relevant parameters of particle swarm.

The relevant parameters comprise population size Pop, acceleration factors c1 and c2, and initial inertia weight omega₀Final inertial weight ω₁Parameters such as maximum iteration times MAXT, the number N of objective functions, the variation probability Mu and the like.

Step 2, based on the given sequencing learning data set, under the ideal sequencing model I, calculating each query q_iBest effectiveness of

And 3, initializing the relevant information of each particle to generate an initial sequencing model set P.

① randomly initialize the position P [ i ] of each particle.

Real number coding is adopted, and in a feasible ordering model domain of an ordering learning problem, an initial position P [ i ] of each particle is randomly generated, namely a weight corresponding to each ordering characteristic, wherein i is more than or equal to 1 and less than or equal to Pop.

② initializes the velocity V [ i ] of each particle to 0.

③ calculate the effectiveness deviation function and the robustness variance function value of the ranking model.

According to the position P [ i ] of each particle]And a linear ranking scoring functionComputing queries q_iEach document d of_ijWherein f is_ijm(q_i,d_ij) Representing query-document pairs (q)_i,d_ij) The mth dimension of (1) ordering the features. According to different Score (q)_i,d_ij) Value-from-big to-little for each query q_iLower documents d_ijCarrying out top-n rapid sequencing, and marking Y according to the sequencing position and the relevance of the document_iCombining the ideal ordering model I, respectively calculating the query q according to the formula (1) to the formula (4)_iAnd the validity deviation function Bias of the query set Q^R(q_i) and Bias^R(Q) and a robust Variance function Variance^R(q_i) And Variance^R(Q) to obtain target values for a validity deviation function and a robustness variance function of the ranking model.

④ initialize the individual extreme value Pbest [ i ] ═ P [ i ] of the particle.

⑤, determining the global extreme value Gbest of the initial population according to the Pbest [ i ] of each particle.

And 4, initializing an iteration counter t to be 0.

And 5, creating an initial sequencing model filing solution set Archive.

In the initial ranking model P, a non-dominant ranking model is selected and stored in a ranking model Archive solution set Archive.

And 6, calculating the congestion distance of each non-dominant solution in the Archive solution set Archive of the sequencing model.

And 7, arranging the non-dominant solutions in Archive in descending order according to the congestion distance.

And 8, performing the following operation on each particle to update information such as the position and the speed of the particle.

①, randomly selecting a certain particle i from the front non-inferior solution set with large crowding distance in the ordered sequencing model Archive solution set Archive, and setting the position of the certain particle i as a global extremum Gbest.

② update the velocity V [ i ] of particle i according to equation (6):

V_im(t+1)＝ω_t*V_im(t)+c1*rand()*[X_im(t)-P_im(t)]+c2*rand()*[X_Gm(t)-P_im(t)]… (6) wherein the inertial weight at the t-th iterationc1 and c2 are acceleration constant factors, and rand () is [0, 1 ]]Random number of cells, X_im(t) and X_Gm(t) respectively representing individual extreme values Pbest [ i ] of the particle i at the t-th iteration]And the m-th dimension component of the global extremum Gbest, wherein the integer i is the particle number and takes the values of 1,2, … … and Pop.

③ updating the position P [ i ] of particle i according to equation (7):

P_im(t+1)＝P_im(t)+V_im(t+1)…(7)

④ checks whether P [ i ] is within the bounds given by the variable and if it is out of its position range, sets the corresponding dimensional variable in P [ i ] to the corresponding bound value with the velocity set to the inverse, i.e., -V [ i ].

⑤ if t < MAXT Mu, then performing mutation operation on the position P [ i ] of the particle with Mu as the mutation rate.

⑥ calculating the objective function value of the particle.

According to the position P [ i ] of the particle]And a linear ranking scoring functionComputing queries q_iEach document d of_ijWherein f is_ijm(q_i,d_ij) Representing query-document pairs (q)_i,d_ij) The mth dimension of (1) ordering the features. According to different Score (q)_i,d_ij) Value-from-big to-little for each query q_iLower documents d_ijCarrying out top-n rapid sequencing, and marking Y according to the sequencing position and the relevance of the document_iCombining the ideal ordering model I, respectively calculating the query q according to the formula (1) to the formula (4)_iAnd the validity deviation function Bias of the query set Q^R(q_i) and Bias^R(Q) and a robust Variance function Variance^R(q_i) And Variance^R(Q) to obtain target values for a validity deviation function and a robustness variance function of the ranking model.

And 9, updating the archiving solution set Archive of the sequencing model.

If the particles in the population P are not dominated by any particle in the order model Archive solution set Archive, then insert all new non-dominated particles in the order model Archive solution set Archive and delete all particles in the order model Archive solution set Archive that are dominated by the new particle. If the Archive solution set of the ordering model is full, replacing the particles according to the following steps:

① calculating the congestion distance of each non-dominant solution in the Archive solution set Archive and arranging in descending order according to the size of the congestion distance.

② randomly selects one particle in the non-dominant solution at the front end with smaller crowding distance from the bottom end of the sequencing model Archive solution set Archive to replace it with a new particle.

And step 10, updating the individual extreme value Pbest [ i ] of each particle in the particle swarm P.

And comparing the new position of the particle P [ i ] with the advantages and disadvantages of the Pbest [ i ] according to the dominance relation, and updating the individual optimum when the P [ i ] is dominant, namely Pbest [ i ] ═ P [ i ].

And 11, if t is t +1, if t is less than MAXT, turning to the step 6, otherwise, outputting each sequencing model in the sequencing model Archive solution set Archive, namely generating a final sequencing model set.

And a third stage: preference of ranking model.

By the idea of a preference order structure assessment method PROMETHEE II based on a multi-attribute decision theory, the effectiveness deviation function Bias of each sequencing model is combined^R(Q) and the robust Variance function Variance^RAnd (Q), selecting a Pareto optimal solution with the maximum net flow sorting value from the sorting model filing solution set generated in the last stage, namely, an optimal sorting model, and taking the Pareto optimal solution as a trained final sorting model. SortingThe preferred implementation flow of the model is shown in fig. 4, and the implementation steps are as follows:

step 1, calculating an outflow function of each sequencing model.

For the ranking model filing solution set Archive finally generated in the second stage, respectively calculating the 'outflow' function Out (R) of each ranking model_i) Value, will Out (R)_i) Is defined as:

wherein | A | represents the total number of Pareto optimized solutions in the ranking model Archive solution set Archive.

And 2, calculating an inflow function of each sequencing model.

For the Archive solution set Archive of the ranking models finally generated In the second stage, the 'inflow' function In (R) of each ranking model is calculated respectively_i) Value In (R)_i) Is defined as:

and 3, calculating a 'net flow' function of each sequencing model.

For the Archive solution set Archive of the ranking models finally generated in the second stage, respectively calculating the Net flow function Net (R) of each ranking model_i) Value, will Net (R)_i) Is defined as:

Net(R_i)＝Out(R_i)-In(R_i)

step 4. according to the Net flow function Net (R)_i) Obtaining each sequencing model R in the sequencing model filing solution set Archive_iThe "net flow" ranking value of (a) and a Pareto optimization solution with the largest "net flow" ranking value is selected as the final ranking model R.

The specific embodiment is as follows:

as shown in fig. 5, assume that there exists a standard rank learning data set L2Rdataset, which is expressed as a set: l2Rdataset { (< q)_i,d_ij＞,y_ij)|q_i∈Q,d_ij∈D_i,y_ij∈Y_i,1≤i≤|Q|,1≤j≤|D_i|}, wherein q_iRepresenting the ith query, Q representing a finite set of queries in the sorted learning data set, | Q | representing the total number of queries in the set of queries, d_ijRepresenting a collection of documents D_iJ document of (1), D_iRepresenting associations to queries q_iDocument collection, | D_iI represents a document set D_iTotal number of documents in, y_ijRepresenting relevance annotations, reflecting the query q_iAnd document d_ijThe degree of correlation between the two groups can be some grade values, such as {1, 2, 3, 4, 5}, Y_i＝{y_i1,y_i2,…,y_i|Di|Denotes the set of tags associated with the query. Query-document pair < q_i,d_ijIs described by an M-dimensional ordering feature f, which can be expressed as < q_i,d_ij＞＝{q_i,f_ij1,f_ij1,…,f_ijm,…,f_ijM}. Through the construction of the "ranking model optimization performance index" of the first stage, four performance indexes are generated: bias^R(q_i)、Bias^R(Q)、Variance^R(q_i) And Variance^R(Q), wherein Bias^R(Q) and Variance^R(Q) is an objective function to be optimized by the multi-objective particle swarm optimization algorithm in the second stage. In the second stage "training of ranking model", for a given ranked learning data set, L2Rdataset, the query-document pairs < q in L2Rdataset are read_i,d_ijCharacteristic value of q_i,f_ij1,f_ij1,…,f_ijm,…,f_ijMAnd relevance notation y_ijWaiting for data, and calculating the Bias under an initial ranking model R^R(q_i)、Bias^R(Q)、Variance^R(q_i) And Variance^R(Q) value, effectiveness deviation function Bias of iterative optimization sequencing model based on multi-objective particle swarm optimization algorithm framework^R(Q) and the robust Variance function Variance^R(Q) to continuously update the ranking model, ultimately producing an Archive solution set Archive of the ranking model. In the third stage, "optimization of the ranking model", for the ranking model filing solution set Archive generated in the previous stage, the "outflow" function value, "inflow" function value and "net flow" function value of each ranking model are calculated according to the corresponding method respectively based on the idea of preference order structure assessment method promete II in the multi-attribute decision theory, and then a Pareto optimization solution with the maximum net flow ranking value, i.e., the optimal ranking model, is selected from the Archive solution set to serve as the trained final ranking model R, the ranking model is a robustness-aware ranking model, and the ranking model can be used to predict the document ranking of new queries. When a new query q needs to retrieve the documents, screening out a corresponding document set in a document database D through a sorting system, scoring the fed-back document set based on a generated sorting model R, then performing descending order on the document set according to the scoring size, generating a sorting result of the predicted documents of the new query q, and outputting and displaying the sorting result.

The second embodiment is as follows:

as shown in fig. 6, the robustness ranking learning method based on multi-objective particle swarm optimization can be applied to the application scenarios of actual demand ranking such as information retrieval, search engines, recommendation systems, question answering systems, and the like. Here, the robust ranking learning method based on multi-objective particle swarm optimization is applied to search engines such as Baidu (Baidu), Google (Google), Bing (Bing), Saogu (Sogou), Yahoo (Yahoo), etc. as application examples. The ranking model trained by the method is embedded into a ranking system of a search engine, and the ranking model is used for predicting the webpage ranking result of the query word required to be searched by the user, so that the satisfaction degree of the whole user can be improved, and the experience of the user is enhanced.

The operation implementation flow of the robustness ranking learning method based on multi-objective particle swarm optimization applied to the search engine is shown in fig. 6, and the implementation steps are as follows:

step 1, integrating a robustness sequencing learning method based on multi-objective particle swarm optimization into a search engine.

Firstly, data preprocessing is carried out on partial webpages in a search engine webpage index database, and extraction and labeling of ranking features are carried out on the webpages to construct a search engine ranking learning data set.

Secondly, on the constructed sequencing learning data set, a robustness sequencing learning method based on multi-objective particle swarm optimization is used for iteratively training a sequencing model to generate a robustness-perceived sequencing model.

Finally, the generated robustness-aware ranking model is embedded in a ranking system of a search engine.

And 2, executing webpage search and presenting a sequencing result.

In a search engine integrated with a robustness ranking learning method based on multi-objective particle swarm optimization, a user can execute webpage search repeatedly in a circulating mode.

First, a user inputs a query word desired to be searched in a search box of a search engine, and clicks a search.

Secondly, the ranking system of the search engine calls a search engine web page index database to find out all web pages containing the query word, and calculates which web pages should be ranked in front and which web pages should be ranked behind to predict the ranking result of the web page search.

Finally, the web page search ranking results are returned to the "search" page in a manner for presentation to the searching user.

While the invention has been described with reference to specific embodiments, the invention is not limited thereto, and various equivalent modifications and substitutions can be easily made by those skilled in the art within the technical scope of the invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (10)

1. A robustness sequencing learning method based on multi-objective particle swarm optimization is characterized by comprising the following steps:

designing an effective deviation function and a robust variance function of a ranking model based on a deviation-variance balance theory, and constructing two optimization performance indexes of ranking learning;

secondly, training a ranking model by using two targets of an effectiveness deviation function and a robustness variance function of an iterative optimization ranking model on a ranking learning data set based on a multi-target particle swarm optimization algorithm framework, so as to generate a ranking model filing solution set;

and thirdly, based on the idea of a preference order structure evaluation method PROMETHEE II in the multi-attribute decision theory, selecting a Pareto optimal sorting model with the maximum 'net flow' sorting value from the sorting model filing solution set generated in the last step to serve as a trained final sorting model.

2. The method for robustness ranking learning based on multi-objective particle swarm optimization according to claim 1, wherein the effectiveness deviation function and the robustness variance function of the design ranking model are used to construct two optimization performance indexes of ranking learning, specifically:

the effectiveness deviation function and the robustness variance function of the query and the query set under the ranking model are respectively defined as follows:

definitions 1 query q_iOf the validity deviation function Bias^R(q_i) Is defined as:

wherein ,representing a query q_iAll documents D under_iThe best effectiveness achieved under the ideal ranking model I, i.e. all documents correctly ranked,representing all documents D under a query qi_iActual effectiveness, Bias, obtained under the ranking model R^R(q_i) Representing a query q_iDeviation of actual effectiveness under the ranking model R from the optimal effectiveness of the ideal ranking model I;

definition 2. validity deviation function Bias of query set Q^R(Q) is defined as:

among them, Bias^R(Q) represents all queries Q in a set Q under a ranking model R_iIs the average of the validity deviations, | Q | represents the query Q in the query set Q_iThe total number of (2);

definitions 3 query q_iRobust Variance function Variance of (1)^R(q_i) Is defined as:

Variance^R(q_i)＝[Bias^R(q_i)-Bias^R(Q)]²…(3)

wherein, Variance^R(q_i) Representing queries q under a ranking model R_iIs of validity Bias^R(q_i) Validity deviation Bias from query set Q^RA degree of dispersion of (Q);

definition 4. robust Variance function Variance of query set Q^R(Q) is defined as:

wherein, Variance^R(Q) represents all queries Q in a set Q under a ranking model R_iThe mean of the robust variance of (a);

converting the robustness ranking learning problem into a multi-objective optimization problem considering both effectiveness and robustness, and formally describing the robustness ranking learning problem as follows according to the definition of the effectiveness deviation function and the robustness variance function:

Utility(Q)＝{min Bias^R(Q),min Variance^R(Q)}…(5)

namely, in the process of sequencing learning, the effectiveness deviation function Bias is minimized at the same time^R(Q) and the robust Variance function Variance^R(Q) to train a ranking model, for which purpose the optimization performance index Bias based on the above-constructed ranking model^R(Q) and Variance^R(Q) a multi-objective intelligent optimization algorithm, such as a multi-objective particle swarm optimization algorithm, can be used while minimizing the Bias^R(Q) and Variance^RThe value of (Q) is used for achieving the purpose of balancing the effectiveness and the robustness of the optimized sequencing model.

3. The method for learning robust ranking based on multi-objective particle swarm optimization according to claim 2, wherein the framework based on multi-objective particle swarm optimization algorithm trains the ranking model according to two objectives of an effectiveness deviation function and a robust variance function of an iterative optimization ranking model, so as to generate a ranking model filing solution set, specifically:

step 1, initializing relevant parameters of a particle swarm;

step 2, based on the given sequencing learning data set, under the ideal sequencing model I, calculating each query q_iBest effectiveness of

Step 3, initializing the relevant information of each particle to generate an initial sequencing model set P;

step 4, initializing an iteration counter t to be 0;

step 5, establishing an initial sequencing model filing solution set Archive, selecting a non-dominant sequencing model from the initial sequencing model P, and storing the non-dominant sequencing model in the sequencing model filing solution set Archive;

step 6, calculating the congestion distance of each non-dominant solution in the archiving solution set Archive of the sequencing model;

step 7, arranging the non-dominant solutions in Archive in a descending order according to the magnitude of the congestion distance;

step 8, performing operation on each particle to update the position and speed information of the particle;

step 9, updating the archiving solution set Archive of the sequencing model;

step 10, updating individual extreme values Pbest [ i ] of each particle in the particle swarm P;

and 11, if t is t +1, if t is less than MAXT, turning to the step 6, otherwise, outputting each sequencing model in the sequencing model Archive solution set Archive, namely generating a final sequencing model set.

4. The method as claimed in claim 3, wherein the relevant parameters of step 1 include population size Pop, acceleration factors c1 and c2, and initial inertial weight ω₀Final inertial weight ω₁Maximum iteration times MAXT, the number N of objective functions and the variation probability Mu.

5. The method according to claim 3, wherein the initializing the relevant information of each particle to generate the initial ranking model set P in step 3 is specifically:

31) randomly initializing the position P [ i ] of each particle;

real number coding is adopted, and in a feasible ordering model domain of an ordering learning problem, an initial position P [ i ] of each particle is randomly generated, namely a weight corresponding to each ordering characteristic, wherein i is more than or equal to 1 and is less than or equal to Pop;

32) initializing the speed V [ i ] of each particle as 0;

33) calculating an effectiveness deviation function and a robustness variance function value of the sequencing model;

according to the position P [ i ] of each particle]And a linear ranking scoring functionComputing queries q_iEach document d of_ijWherein f is_ijm(q_i,d_ij) Representing query-document pairs (q)_i,d_ij) According to different Score (q)_i,d_ij) Value-from-big to-little for each query q_iLower documents d_ijCarrying out top-n rapid sequencing, and marking Y according to the sequencing position and the relevance of the document_iCombining the ideal ordering model I, respectively calculating the query q according to the formula (1) to the formula (4)_iAnd the validity deviation function Bias of the query set Q^R(q_i) and Bias^R(Q) and a robust Variance function Variance^R(q_i) And Variance^R(Q) to obtain target values for an effectiveness bias function and a robustness variance function of the ranking model;

34) initializing an individual extreme value Pbest [ i ] ═ P [ i ] of the particle;

35) and determining the global extreme value Gbest of the initial population according to the Pbest [ i ] of each particle.

6. The method for robustness ranking learning based on multi-objective particle swarm optimization according to claim 3, wherein the step 8 of performing an operation on each particle to update the position and velocity information of the particle is specifically as follows:

81) randomly selecting a certain particle i from a non-dominant solution set at the front end with a larger crowding distance in the ordered sequencing model filing solution set Archive, and setting the position of the certain particle i as a global extremum Gbest;

82) updating the velocity V [ i ] of particle i according to equation (6):

V_im(t+1)＝ω_t*V_im(t)+c1*rand()*[X_im(t)-P_im(t)]+c2*rand()*[X_Gm(t)-P_im(t)]…(6)

wherein the inertia weight at the t-th iterationc1 and c2 are acceleration constant factors, and rand () is [0, 1 ]]Random number of cells, X_im(t) and X_Gm(t) respectively representing individual extreme values Pbest [ i ] of the particle i at the t-th iteration]The m-th dimension component of Gbest and the m-th dimension component of the global extreme value Gbest, wherein the integer i is the particle number and takes the values of 1,2, … … and Pop;

83) updating the position P [ i ] of the particle i according to equation (7):

P_im(t+1)＝P_im(t)+V_im(t+1)…(7)

84) checking whether P [ i ] is within the limits given by the variables, and if the position range of P [ i ] is exceeded, setting the corresponding dimensional variable in P [ i ] as the corresponding boundary value and setting the speed as the reverse direction, namely-V [ i ];

85) if t < MAXT Mu, then executing variation operation on the position P [ i ] of the particle by taking Mu as variation rate;

86) calculating an objective function value of the particle;

according to the position P [ i ] of the particle]And a linear ranking scoring functionComputing queries q_iEach document d of_ijWherein f is_ijm(q_i,d_ij) Representing query-document pairs (q)_i,d_ij) According to different Score (q)_i,d_ij) Value-from-big to-little for each query q_iLower documents d_ijCarrying out top-n rapid sequencing, and marking Y according to the sequencing position and the relevance of the document_iCombining the ideal ordering model I, respectively calculating the query q according to the formula (1) to the formula (4)_iAnd the validity deviation function Bias of the query set Q^R(q_i) and Bias^R(Q) and a robust Variance function Variance^R(q_i) And Variance^R(Q) to obtain target values for a validity deviation function and a robustness variance function of the ranking model.

7. The method for robustness ranking learning based on multi-objective particle swarm optimization according to claim 3, wherein the step 9 of updating the ranking model Archive solution set Archive specifically comprises:

if the particles in the population P are not dominated by any particle in the order model Archive solution set Archive, inserting all new non-dominated particles in the order model Archive solution set Archive and deleting all particles dominated by the new particles in the order model Archive solution set Archive, if the order model Archive solution set Archive is full, replacing the particles by the following steps:

step 1, calculating the congestion distance of each non-dominant solution in an Archive solution set Archive of a sequencing model, and arranging the congestion distances in a descending order according to the size of the congestion distance;

and 2, randomly selecting one particle in the non-dominant solution at the front end with smaller crowding distance from the bottom end of the sequencing model Archive solution set Archive, and replacing the particle with a new particle.

8. The method for robustness ranking learning based on multi-objective particle swarm optimization according to claim 3, wherein the step 10 of updating the individual extremum Pbest [ i ] of each particle in the particle swarm P is specifically as follows:

and comparing the new position of the particle P [ i ] with the advantages and disadvantages of the Pbest [ i ] according to the dominance relation, and updating the individual optimum when the P [ i ] is dominant, namely Pbest [ i ] ═ P [ i ].

9. The method for learning robustness ranking based on multi-objective particle swarm optimization according to claim 1, wherein in the third step, based on an idea of a preference order structure assessment method promethe II in a multi-attribute decision theory, a Pareto optimal ranking model with a maximum "net flow" ranking value is selected from the ranking model filing solution set generated in the previous step, and the selected ranking model is specifically:

step 1, calculating an outflow function of each sequencing model;

respectively calculating the 'outflow' function Out (R) of each sequencing model for the sequencing model filing solution set Archive finally generated in the step two_i) Value, will Out (R)_i) Is defined as:

wherein, | A | represents the total number of Pareto optimized solutions in the ranking model Archive solution set Archive;

step 2, calculating an inflow function of each sequencing model;

respectively calculating the 'inflow' function In (R) of each sequencing model for the sequencing model filing solution set Archive finally generated In the step two_i) Value In (R)_i) Is defined as:

step 3, calculating a net flow function of each sequencing model;

respectively calculating the Net flow function Net (R) of each sequencing model for the sequencing model filing solution set Archive finally generated in the step two_i) Value, will Net (R)_i) Is defined as:

Net(R_i)＝Out(R_i)-In(R_i)

step 4. according to the Net flow function Net (R)_i) Obtaining each sequencing model R in the sequencing model filing solution set Archive_iThe "net flow" ranking value of (a) and a Pareto optimization solution with the largest "net flow" ranking value is selected as the final ranking model R.

10. The application of the robustness ranking learning method based on multi-objective particle swarm optimization is characterized in that the robustness ranking learning method based on multi-objective particle swarm optimization is applied to a search engine, wherein the search engine comprises Baidu, Google, Bing, Saogouu and Yahoo, a ranking model trained by the method is embedded into a ranking system of the search engine, and the ranking model is used for predicting a webpage ranking result of query words required to be searched by a user, so that the satisfaction degree of the whole user can be improved, and the experience of the user is enhanced, and the specific application process is as follows:

step 1, integrating a robustness sequencing learning method based on multi-objective particle swarm optimization into a search engine;

firstly, performing data preprocessing on partial webpages in a search engine webpage index database, and extracting and labeling ranking features of the webpages to construct a search engine ranking learning data set;

secondly, iteratively training a ranking model to generate a robustness-perceived ranking model by using a robustness ranking learning method based on multi-objective particle swarm optimization on the constructed ranking learning data set;

finally, embedding the generated robustness sensing sequencing model into a sequencing system of a search engine;

step 2, executing web page search and presenting a sequencing result;

in a search engine integrated with a robustness ranking learning method based on multi-objective particle swarm optimization, a user can circularly execute webpage search for multiple times;

firstly, a user inputs a query word to be searched in a search box of a search engine and clicks for searching;

secondly, a search engine webpage index database is called by a search engine ranking system, all webpages containing the query word are found out from the database, and the ranking results of webpage search are predicted by calculating which webpages should be ranked in front and which should be ranked behind;

and finally, returning the webpage search sorting result to a search page according to a set mode for presenting to a search user.