Processing math: 100%
Research article

DyCARS: A dynamic context-aware recommendation system


  • Dynamic recommendation systems aim to achieve real-time updates and dynamic migration of user interests, primarily utilizing user-item interaction sequences with timestamps to capture the dynamic changes in user interests and item attributes. Recent research has mainly centered on two aspects. First, it involves modeling the dynamic interaction relationships between users and items using dynamic graphs. Second, it focuses on mining their long-term and short-term interaction patterns. This is achieved through the joint learning of static and dynamic embeddings for both users and items. Although most existing methods have achieved some success in modeling the historical interaction sequences between users and items, there is still room for improvement, particularly in terms of modeling the long-term dependency structures of dynamic interaction histories and extracting the most relevant delayed interaction patterns. To address this issue, we proposed a Dynamic Context-Aware Recommendation System for dynamic recommendation. Specifically, our model is built on a dynamic graph and utilizes the static embeddings of recent user-item interactions as dynamic context. Additionally, we constructed a Gated Multi-Layer Perceptron encoder to capture the long-term dependency structure in the dynamic interaction history and extract high-level features. Then, we introduced an Attention Pooling network to learn similarity scores between high-level features in the user-item dynamic interaction history. By calculating bidirectional attention weights, we extracted the most relevant delayed interaction patterns from the historical sequence to predict the dynamic embeddings of users and items. Additionally, we proposed a loss function called the Pairwise Cosine Similarity loss for dynamic recommendation to jointly optimize the static and dynamic embeddings of two types of nodes. Finally, extensive experiments on two real-world datasets, LastFM, and the Global Terrorism Database showed that our model achieves consistent improvements over state-of-the-art baselines.

    Citation: Zhiwen Hou, Fanliang Bu, Yuchen Zhou, Lingbin Bu, Qiming Ma, Yifan Wang, Hanming Zhai, Zhuxuan Han. DyCARS: A dynamic context-aware recommendation system[J]. Mathematical Biosciences and Engineering, 2024, 21(3): 3563-3593. doi: 10.3934/mbe.2024157

    Related Papers:

    [1] Bashir Ahmad, Ahmed Alsaedi, Ymnah Alruwaily, Sotiris K. Ntouyas . Nonlinear multi-term fractional differential equations with Riemann-Stieltjes integro-multipoint boundary conditions. AIMS Mathematics, 2020, 5(2): 1446-1461. doi: 10.3934/math.2020099
    [2] Ymnah Alruwaily, Lamya Almaghamsi, Kulandhaivel Karthikeyan, El-sayed El-hady . Existence and uniqueness for a coupled system of fractional equations involving Riemann-Liouville and Caputo derivatives with coupled Riemann-Stieltjes integro-multipoint boundary conditions. AIMS Mathematics, 2023, 8(5): 10067-10094. doi: 10.3934/math.2023510
    [3] Subramanian Muthaiah, Dumitru Baleanu, Nandha Gopal Thangaraj . Existence and Hyers-Ulam type stability results for nonlinear coupled system of Caputo-Hadamard type fractional differential equations. AIMS Mathematics, 2021, 6(1): 168-194. doi: 10.3934/math.2021012
    [4] Bashir Ahmad, Ahmed Alsaedi, Areej S. Aljahdali, Sotiris K. Ntouyas . A study of coupled nonlinear generalized fractional differential equations with coupled nonlocal multipoint Riemann-Stieltjes and generalized fractional integral boundary conditions. AIMS Mathematics, 2024, 9(1): 1576-1594. doi: 10.3934/math.2024078
    [5] Md. Asaduzzaman, Md. Zulfikar Ali . Existence of positive solution to the boundary value problems for coupled system of nonlinear fractional differential equations. AIMS Mathematics, 2019, 4(3): 880-895. doi: 10.3934/math.2019.3.880
    [6] Murugesan Manigandan, Kannan Manikandan, Hasanen A. Hammad, Manuel De la Sen . Applying fixed point techniques to solve fractional differential inclusions under new boundary conditions. AIMS Mathematics, 2024, 9(6): 15505-15542. doi: 10.3934/math.2024750
    [7] Kishor D. Kucche, Sagar T. Sutar, Kottakkaran Sooppy Nisar . Analysis of nonlinear implicit fractional differential equations with the Atangana-Baleanu derivative via measure of non-compactness. AIMS Mathematics, 2024, 9(10): 27058-27079. doi: 10.3934/math.20241316
    [8] Subramanian Muthaiah, Manigandan Murugesan, Muath Awadalla, Bundit Unyong, Ria H. Egami . Ulam-Hyers stability and existence results for a coupled sequential Hilfer-Hadamard-type integrodifferential system. AIMS Mathematics, 2024, 9(6): 16203-16233. doi: 10.3934/math.2024784
    [9] Bashir Ahmad, Manal Alnahdi, Sotiris K. Ntouyas, Ahmed Alsaedi . On a mixed nonlinear boundary value problem with the right Caputo fractional derivative and multipoint closed boundary conditions. AIMS Mathematics, 2023, 8(5): 11709-11726. doi: 10.3934/math.2023593
    [10] Djamila Chergui, Taki Eddine Oussaeif, Merad Ahcene . Existence and uniqueness of solutions for nonlinear fractional differential equations depending on lower-order derivative with non-separated type integral boundary conditions. AIMS Mathematics, 2019, 4(1): 112-133. doi: 10.3934/Math.2019.1.112
  • Dynamic recommendation systems aim to achieve real-time updates and dynamic migration of user interests, primarily utilizing user-item interaction sequences with timestamps to capture the dynamic changes in user interests and item attributes. Recent research has mainly centered on two aspects. First, it involves modeling the dynamic interaction relationships between users and items using dynamic graphs. Second, it focuses on mining their long-term and short-term interaction patterns. This is achieved through the joint learning of static and dynamic embeddings for both users and items. Although most existing methods have achieved some success in modeling the historical interaction sequences between users and items, there is still room for improvement, particularly in terms of modeling the long-term dependency structures of dynamic interaction histories and extracting the most relevant delayed interaction patterns. To address this issue, we proposed a Dynamic Context-Aware Recommendation System for dynamic recommendation. Specifically, our model is built on a dynamic graph and utilizes the static embeddings of recent user-item interactions as dynamic context. Additionally, we constructed a Gated Multi-Layer Perceptron encoder to capture the long-term dependency structure in the dynamic interaction history and extract high-level features. Then, we introduced an Attention Pooling network to learn similarity scores between high-level features in the user-item dynamic interaction history. By calculating bidirectional attention weights, we extracted the most relevant delayed interaction patterns from the historical sequence to predict the dynamic embeddings of users and items. Additionally, we proposed a loss function called the Pairwise Cosine Similarity loss for dynamic recommendation to jointly optimize the static and dynamic embeddings of two types of nodes. Finally, extensive experiments on two real-world datasets, LastFM, and the Global Terrorism Database showed that our model achieves consistent improvements over state-of-the-art baselines.



    The topic of fractional differential equations received immense popularity and attraction due to their extensive use in the mathematical modeling of several real world phenomena. Examples include HIV-immune system with memory [1], stabilization of chaotic systems [2], chaotic synchronization [3,4], ecology [5], infectious diseases [6], economic model [7], fractional neural networks [8,9], COVID-19 infection [10], etc. A salient feature distinguishing fractional-order differential and integral operators from the classical ones is their nonlocal nature, which can provide the details about the past history of the phenomena and processes under investigation. In the recent years, many researchers contributed to the development of fractional calculus, for example, see [11,12,13,14,15,16,17,18,19,20,21,22,23,24] and the references cited therein. One can also find a substantial material about fractional order coupled systems in the articles [25,26,27,28,29,30,31,32,33,34].

    In this paper, motivated by [30], we consider a Caputo type coupled system of nonlinear fractional differential equations supplemented with a new set of boundary conditions in terms of the sum and difference of the governing functions given by

    {CDνφ(t)=f(t,φ(t),ψ(t)),tJ:=[0,T],CDρψ(t)=g(t,φ(t),ψ(t)),tJ:=[0,T],P1(φ+ψ)(0)+P2(φ+ψ)(T)=mi=1ai(φ+ψ)(σi),T0(φψ)(s)dsζη(φψ)(s)ds=A, (1.1)

    where CDχ is the Caputo fractional derivative operator of order χ{ν,ρ}, ν,ρ(0,1], 0<σi<η<ζ<T, i=1,,m (the case 0<η<ζ<σi<T can be treated in a similar way), P1,P2,ai,A are nonnegative constants, such that P1+P2mi=1ai0, Tζ+η0, and f,g:[0,T]×R2R are continuous functions.

    Here it is imperative to notice that the first condition introduced in the problem (1.1) can be interpreted as the sum of the governing functions φ and ψ at the end positions of the interval [0,T] is sum of similar contributions due to arbitrary positions at σi(0,T),i=1,...,m, while the second condition describes that the contribution of the difference of the governing functions φ and ψ on the domain [0,T] differs from the one an arbitrary sub-domain (η,ξ) by a constant A.

    We will also study the problem (1.1) by replacing A in the last condition with the one containing nonlinear Riemann-Liouville integral term of the form:

    1Γ(δ)T0(Ts)δ1h(s,φ(s),ψ(s))ds,δ>0, (1.2)

    where h:[0,T]×R2R is a given continuous function.

    We organize the rest of the paper as follows. In Section 2, we outline the related concepts of fractional calculus and establish an auxiliary lemma for the linear analogue of the problem (1.1). We apply the standard fixed point theorems to derive the existence and uniqueness results for the problem (1.1) in Section 3. The case of nonlinear Riemann-Liouville integral boundary conditions is discussed in Section 4. The paper concludes with some interesting observations and special cases.

    Let us begin this section with some preliminary concepts of fractional calculus [11].

    Definition 2.1. The Riemann-Liouville fractional integral of order q>0 of a function h:[0,)R is defined by

    Iqh(t)=t0(ts)q1Γ(q)h(s)ds,t>0,

    provided the right-hand side is point-wise defined on (0,), where Γ is the Gamma function.

    Definition 2.2. The Caputo fractional derivative of order q for a function h:[0,]R with h(t)ACn[0,) is defined by

    CDqh(t)=1Γ(nq)t0h(n)(s)(ts)qn+1ds=Inqh(n)(t), t>0,n1<q<n.

    Lemma 2.1. Let q>0 and h(t)ACn[0,) or Cn[0,). Then

    (IqCDqh)(t)=h(t)n1k=0h(k)(0)k!tk,t>0,n1<q<n. (2.1)

    Now we present an auxiliary lemma related to the linear variant of problem (1.1).

    Lemma 2.2. Let F,GC[0,T], φ,ψAC[0,T].Then the solution of the following linear coupled system:

    {CDνφ(t)=F(t),tJ:=[0,T],CDρψ(t)=G(t),tJ:=[0,T],P1(φ+ψ)(0)+P2(φ+ψ)(T)=mi=1ai(φ+ψ)(σi),T0(φψ)(s)dsζη(φψ)(s)ds=A, (2.2)

    is given by

    φ(t)=t0(ts)ν1Γ(ν)F(s)ds+12{AΛ21Λ2T0(s0(sx)ν1Γ(ν)F(x)dxs0(sx)ρ1Γ(ρ)G(x)dx)dsP2Λ1(T0(Ts)ν1Γ(ν)F(s)ds+T0(Ts)ρ1Γ(ρ)G(s)ds)+1Λ2ξη(s0(sx)ν1Γ(ν)F(x)dxs0(sx)ρ1Γ(ρ)G(x)dx)ds+mi=1aiΛ1(σi0(σis)ν1Γ(ν)F(s)ds+σi0(σis)ρ1Γ(ρ)G(s)ds)}, (2.3)
    ψ(t)=t0(ts)ρ1Γ(ρ)G(s)ds+12{AΛ2+1Λ2T0(s0(sx)ν1Γ(ν)F(x)dxs0(sx)ρ1Γ(ρ)G(x)dx)dsP2Λ1(T0(Ts)ν1Γ(ν)F(s)ds+T0(Ts)ρ1Γ(ρ)G(s)ds)1Λ2ξη(s0(sx)ν1Γ(ν)F(x)dxs0(sx)ρ1Γ(ρ)G(x)dx)ds+mi=1aiΛ1(σi0(σis)ν1Γ(ν)F(s)ds+σi0(σis)ρ1Γ(ρ)G(s)ds)}, (2.4)

    where

    Λ1:=P1+P2mi=1ai0, (2.5)
    Λ2:=Tζ+η0. (2.6)

    Proof. Applying the operators Iν and Iρ on the first and second fractional differential equations in (2.2) respectively and using Lemma 2.1, we obtain

    φ(t)=t0(ts)ν1Γ(ν)F(s)ds+c1, (2.7)
    ψ(t)=t0(ts)ρ1Γ(ρ)G(s)ds+c2, (2.8)

    where c1,c2R. Inserting (2.7) and (2.8) in the condition P1(φ+ψ)(0)+P2(φ+ψ)(T)=mi=1ai(φ+ψ)(σi), we get

    c1+c2=1Λ1{mi=1ai(σi0(σis)ν1Γ(ν)F(s)ds+σi0(σis)ρ1Γ(ρ)G(s)ds)P2(T0(Ts)ν1Γ(ν)F(s)ds+T0(Ts)ρ1Γ(ρ)G(s)ds)}. (2.9)

    Using (2.7) and (2.8) in the condition T0(φψ)(s)dsζη(φψ)(s)ds=A, we obtain

    c1c2=1Λ2{AT0(s0(sx)ν1Γ(ν)F(x)dxs0(sx)ρ1Γ(ρ)G(x)dx)ds+ξη(s0(sx)ν1Γ(ν)F(x)dxs0(sx)ρ1Γ(ρ)G(x)dx)ds}. (2.10)

    Solving (2.9) and (2.10) for c1 and c2, yields

    c1=12{AΛ21Λ2T0(s0(sx)ν1Γ(ν)F(x)dxs0(sx)ρ1Γ(ρ)G(x)dx)dsP2Λ1(T0(Ts)ν1Γ(ν)F(s)ds+T0(Ts)ρ1Γ(ρ)G(s)ds)+1Λ2ξη(s0(sx)ν1Γ(ν)F(x)dxs0(sx)ρ1Γ(ρ)G(x)dx)ds+1Λ1mi=1ai(σi0(σis)ν1Γ(ν)F(s)ds+σi0(σis)ρ1Γ(ρ)G(s)ds)},

    and

    c2=12{AΛ2+1Λ2T0(s0(sx)ν1Γ(ν)F(x)dxs0(sx)ρ1Γ(ρ)G(x)dx)dsP2Λ1(T0(Ts)ν1Γ(ν)F(s)ds+T0(Ts)ρ1Γ(ρ)G(s)ds)1Λ2ξη(s0(sx)ν1Γ(ν)F(x)dxs0(sx)ρ1Γ(ρ)G(x)dx)ds+1Λ1mi=1ai(σi0(σis)ν1Γ(ν)F(s)ds+σi0(σis)ρ1Γ(ρ)G(s)ds)}.

    Substituting the values of c1 and c2 in (2.7) and (2.8) respectively, we get the solution (2.3) and (2.4). By direct computation, one can obtain the converse of this lemma. The proof is complete.

    Let X=C([0,T],R)×C([0,T],R) denote the Banach space endowed with the norm (φ,ψ)=φ+ψ=supt[0,T]|φ(t)|+supt[0,T]|ψ(t)|, (φ,ψ)X. In view of Lemma 2.2, we define an operator Φ:XX in relation to the problem (1.1) as

    Φ(φ,ψ)(t):=(Φ1(φ,ψ)(t),Φ2(φ,ψ)(t)), (3.1)

    where

    Φ1(φ,ψ)(t)=1Γ(ν)t0(ts)ν1f(s,φ(s),ψ(s))ds+12{AΛ21Λ2T0(s0(sx)ν1Γ(ν)f(x,φ(x),ψ(x))dxs0(sx)ρ1Γ(ρ)g(x,φ(x),ψ(x))dx)dsP2Λ1(T0(Ts)ν1Γ(ν)f(s,φ(s),ψ(s))ds+T0(Ts)ρ1Γ(ρ)g(s,φ(s),ψ(s))ds)+1Λ2ξη(s0(sx)ν1Γ(ν)f(x,φ(x),ψ(x))dxs0(sx)ρ1Γ(ρ)g(x,φ(x),ψ(x))dx)ds+1Λ1mi=1ai(σi0(σis)ν1Γ(ν)f(s,φ(s),ψ(s))ds+σi0(σis)ρ1Γ(ρ)g(s,φ(s),ψ(s))ds)}, (3.2)

    and

    Φ2(φ,ψ)(t)=1Γ(ρ)t0(ts)ρ1g(s,φ(s),ψ(s))ds+12{AΛ2+1Λ2T0(s0(sx)ν1Γ(ν)f(x,φ(x),ψ(x))dxs0(sx)ρ1Γ(ρ)g(x,φ(x),ψ(x))dx)dsP2Λ1(T0(Ts)ν1Γ(ν)f(s,φ(s),ψ(s))ds+T0(Ts)ρ1Γ(ρ)g(s,φ(s),ψ(s))ds)1Λ2ξη(s0(sx)ν1Γ(ν)f(x,φ(x),ψ(x))dxs0(sx)ρ1Γ(ρ)g(x,φ(x),ψ(x))dx)ds+1Λ1mi=1ai(σi0(σis)ν1Γ(ν)f(s,φ(s),ψ(s))ds+σi0(σis)ρ1Γ(ρ)g(s,φ(s),ψ(s))ds)}. (3.3)

    In the forthcoming analysis, we need the following assumptions.

    (H1) There exist continuous nonnegative functions μi,κiC([0,1],R+),i=1,2,3, such that

    |f(t,φ,ψ)|μ1(t)+μ2(t)|φ|+μ3(t)|ψ|(t,φ,ψ)J×R2;
    |g(t,φ,ψ)|κ1(t)+κ2(t)|φ|+κ3(t)|ψ|(t,φ,ψ)J×R2.

    (H2) There exist positive constants αi,βi,i=1,2, such that

    |f(t,φ1,ψ1)f(t,φ2,ψ2)|α1|φ1φ2|+α2|ψ1ψ2|,tJ,φi,ψiR,i=1,2;
    |g(t,φ1,ψ1)g(t,φ2,ψ2)|β1|φ1φ2|+β2|ψ1ψ2|,tJ,φi,ψiR,i=1,2.

    For computational convenience, we introduce the notation:

    ϱ1=12|Λ1|[mi=1aiσνiΓ(ν+1)+P2TνΓ(ν+1)]+12|Λ2|[ζν+1ην+1Γ(ν+2)+Tν+1Γ(ν+2)], (3.4)
    ϱ2=12|Λ1|[mi=1aiσρiΓ(ρ+1)+P2TρΓ(ρ+1)]+12|Λ2|[ζρ+1ηρ+1Γ(ρ+2)+Tρ+1Γ(ρ+2)], (3.5)

    and

    M0=min{1[μ2(2ϱ1+TνΓ(ν+1))+κ2(2ϱ2+TρΓ(ρ+1))],1[μ3(2ϱ1+TνΓ(ν+1))+κ3(2ϱ2+TρΓ(ρ+1))]}.

    We make use of the following fixed point theorem [35] to prove the existence of solutions for the problem (1.1).

    Lemma 3.1. Let E be the Banach space and Q:EE be a completely continuous operator. If the set Ω={xE|x=μQx,0<μ<1} is bounded, then Q has a fixed point in E.

    Theorem 3.1. Suppose that f,g:J×R2R are continuousfunctions and the condition (H1) holds. Then there exists at least one solution for the problem (1.1) on J if

    μ2(2ϱ1+TνΓ(ν+1))+κ2(2ϱ2+TρΓ(ρ+1))<1,μ3(2ϱ1+TνΓ(ν+1))+κ3(2ϱ2+TρΓ(ρ+1))<1, (3.6)

    where ϱi(i=1,2) are defined in (3.4)–(3.5).

    Proof. Observe that continuity of Φ:XX follows from that of the functions f and g. Now we show that the operator Φ maps any bounded subset of X into a relatively compact subset of X. For that, let ΩˉrX be bounded. Then, for the positive real constants Lf and Lg, we have

    |f(t,φ(t),ψ(t))|Lf,|g(t,φ(t),ψ(t))|Lg,(φ,ψ)Ωˉr.

    So, for any (φ,ψ)Ωˉr, tJ, we get

    |Φ1(φ,ψ)(t)|LfΓ(ν)t0(ts)ν1ds+12{A|Λ2|+1Λ2T0(Lfs0(sx)ν1Γ(ν)dx+Lgs0(sx)ρ1Γ(ρ)dx)ds+P2|Λ1|(LfT0(Ts)ν1Γ(ν)ds+LgT0(Ts)ρ1Γ(ρ)ds)+1|Λ2|ξη(Lfs0(sx)ν1Γ(ν)dx+Lgs0(sx)ρ1Γ(ρ)dx)ds+1|Λ1|mi=1ai(Lfσi0(σis)ν1Γ(ν)ds+Lgσi0(σis)ρ1Γ(ρ)ds)}LfTνΓ(ν+1)+Lf2|Λ1|[mi=1aiσνiΓ(ν+1)+P2TνΓ(ν+1)]+Lf2|Λ2|[ζν+1ην+1Γ(ν+2)+Tν+1Γ(ν+2)]+Lg2|Λ1|[mi=1aiσρiΓ(ρ+1)+P2TρΓ(ρ+1)]+Lg2|Λ2|[ζρ+1ηρ+1Γ(ρ+2)+Tρ+1Γ(ρ+2)]+A2|Λ2|,

    which, in view of (3.4) and (3.5), takes the form:

    |Φ1(φ,ψ)(t)|Lf(TνΓ(ν+1)+ϱ1)+Lgϱ2+A2|Λ2|. (3.7)

    In a similar fashion, one can obtain

    |Φ2(φ,ψ)(t)|Lfϱ1+Lg(TρΓ(ρ+1)+ϱ2)+A2|Λ2|. (3.8)

    From (3.7) and (3.8), we get

    Φ(φ,ψ)=Φ1(φ,ψ)+Φ2(φ,ψ)Lf(TνΓ(ν+1)+2ϱ1)+Lg(TρΓ(ρ+1)+2ϱ2)+A|Λ2|.

    From the foregoing inequality, we deduce that the operator Φ is uniformly bounded.

    In order to show that Φ maps bounded sets into equicontinuous sets of X, let t1,t2[0,T],t1<t2, and (φ,ψ)Ωˉr. Then

    |Φ1(φ,ψ)(t2)Φ1(φ,ψ)(t1)||1Γ(ν)(t10[(t2s)ν1(t1s)ν1]f(s,φ(s),ψ(s))ds+t2t1(t2s)ν1f(s,φ(s),ψ(s))ds)|Lf(2(t2t1)ν+tν2tν1Γ(ν+1)).

    Analogously, we can obtain

    |Φ2(φ,ψ)(t2)Φ2(u,v)(t1)|Lg(2(t2t1)ρ+tρ2tρ1Γ(ρ+1)).

    Clearly the right-hand sides of the above inequalities tend to zero when t1t2, independently of (φ,ψ)Ωˉr. Thus it follows by the Arzelá-Ascoli theorem that the operator Φ:XX is completely continuous.

    Next we consider the set E={(φ,ψ)X|(φ,ψ)=λΦ(φ,ψ),0<λ<1} and show that it is bounded. Let (φ,ψ)E, then (φ,ψ)=λΦ(φ,ψ),0<λ<1. For any tJ, we have

    φ(t)=λΦ1(φ,ψ)(t),ψ(t)=λΦ2(φ,ψ)(t).

    As in the previous step, using ϱi(i=1,2) given by (3.4)-(3.5), we find that

    |φ(t)|=λ|Φ1(φ,ψ)(t)|(μ1+μ2φ+μ3ψ)(TνΓ(ν+1)+ϱ1)+(κ1+κ2φ+κ3ψ)ϱ2+A2|Λ2|,
    |ψ(t)|=λ|Φ2(φ,ψ)(t)|(μ1+μ2φ+μ3ψ)ϱ1+(κ1+κ2φ+κ3ψ)(TρΓ(ρ+1)+ϱ2)+A2|Λ2|.

    In consequence, we get

    φ+ψμ1(2ϱ1+TνΓ(ν+1))+κ1(2ϱ2+TρΓ(ρ+1))+A|Λ2|+[μ2(2ϱ1+TνΓ(ν+1))+κ2(2ϱ2+TρΓ(ρ+1))]φ+[μ3(2ϱ1+TνΓ(ν+1))+κ3(2ϱ2+TνΓ(ν+1))]ψ.

    Thus, by the condition (3.6), we have

    (φ,ψ)1M0{μ1(2ϱ1+TνΓ(ν+1))+κ1(2ϱ2+TρΓ(ρ+1))+A|Λ2|},

    which shows that (φ,ψ) is bounded for tJ. In consequence, the set E is bounded. Thus it follows by the conclusion of Lemma 3.1 that the operator Φ has at least one fixed point, which is indeed a solution of the problem (1.1).

    Letting μ2(t)=μ3(t)0 and κ2(t)=κ3(t)0, the statement of Theorem 3.1 takes the following form.

    Corollary 3.1. Let f,g:J×R2R be continuousfunctions such that

    |f(t,φ,ψ)|μ1(t),|g(t,φ,ψ)|κ1(t),(t,φ,ψ)J×R2,

    where μ1,κ1C([0,T],R+). Then there exists at least one solution for the problem (1.1) on J.

    Corollary 3.2. If μi(t)=λi,κi(t)=εi,i=1,2,3, then the condition (H1) becomes:

    (H1) there exist real constants λi,εi>0,i=1,2, such that

    |f(t,φ,ψ)|λ1+λ2|φ|+λ3|ψ|(t,φ,ψ)J×R2;
    |f(t,φ,ψ)|ε1+ε2|φ|+ε3|ψ|(t,φ,ψ)J×R2;

    and (3.6) takes the form:

    λ2(2ϱ1+TνΓ(ν+1))+ε2(2ϱ2+TρΓ(ρ+1))<1,λ3(2ϱ1+TνΓ(ν+1))+ε3(2ϱ2+TρΓ(ρ+1))<1.

    Then there exists at least one solution for the problem (1.1) on J.

    The next result is concerned with the existence of a unique solution for the problem (1.1) and is reliant on the contraction mapping principle due to Banach.

    Theorem 3.2. Let f,g:[0,1]×R2R be continuous functions and the assumption (H2) holds.Then the problem (1.1) has a unique solution on J if

    α(TνΓ(ν+1)+2ϱ1)+β(TρΓ(ρ+1)+2ϱ2)<1, (3.9)

    where α=max{α1,α2},β=max{β1,β2} and ϱi,i=1,2, are defined in (3.4)-(3.5).

    Proof. Consider the operator Φ:XX defined by (3.1) and take

    r>M1(TνΓ(ν+1)+2ϱ1)+M2(TρΓ(ρ+1)+2ϱ2)+A|Λ2|1(α(TνΓ(ν+1)+2ϱ1)+β(TρΓ(ρ+1)+2ϱ2)),

    where M1=supt[0,T]|f(t,0,0)|, and M2=supt[0,T]|g(t,0,0)|. Then we show that ΦBrBr, where Br={(φ,ψ)X:(φ,ψ)r}. By the assumption (H1), for (φ,ψ)Br,t[0,T], we have

    |f(t,φ(t),ψ(t))||f(t,φ(t),ψ(t))f(t,0,0)|+|f(t,0,0)|α(|φ(t)|+|ψ(t)|)+M1α(φ+ψ)+M1.

    In a similar manner, one can find that

    |g(t,φ(t),ψ(t))|β(φ+ψ)+M2.

    In consequence, for (φ,ψ)Br, we obtain

    |Φ1(φ,ψ)(t)|TνΓ(ν+1)(α(φ+ψ)+M1)+12[A|Λ2|+1|Λ2|(Tν+1Γ(ν+2)(α(φ+ψ)+M1)+Tρ+1Γ(ρ+2)(β(φ+ψ)+M2))+P2|Λ1|(TνΓ(ν+1)(α(φ+ψ)+M1)+TρΓ(ρ+1)(β(φ+ψ)+M2))+1|Λ2|(ζν+1ην+1Γ(ν+2)(α(φ+ψ)+M1)+ζρ+1ηρ+1Γ(ρ+2)(β(φ+ψ)+M2))+1|Λ1|mi=1ai(σνiΓ(ν+1)(α(φ+ψ)+M1)+σρiΓ(ρ+1)(β(φ+ψ)+M2))],

    which, on taking the norm for tJ, yields

    Φ1(φ,ψ)(α(TνΓ(ν+1)+ϱ1)+βϱ2)(φ+ψ)+M1(TνΓ(ν+1)+ϱ1)+M2ϱ2+A2|Λ2|.

    In the same way, for (φ,ψ)Br, one can obtain

    Φ2(φ,ψ)(αϱ1+β(TρΓ(ρ+1)+ϱ2))(φ+ψ)+M1ϱ1+M2(TρΓ(ρ+1)+ϱ2)+A2|Λ2|.

    Therefore, for any (φ,ψ)Br, we have

    Φ(φ,ψ))=Φ1(φ,ψ)+Φ2(φ,ψ)(α(TνΓ(ν+1)+2ϱ1)+β(TρΓ(ρ+1)+2ϱ2))(φ+ψ)+M1(TνΓ(ν+1)+2ϱ1)+M2(TρΓ(ρ+1)+2ϱ2)+A|Λ2|<r,

    which shows that Φ maps Br into itself.

    Next it will be shown that the operator Φ is a contraction. For (φ1,ψ1),(φ2,ψ2)E,t[0,T], it follows by (H2) that

    |Φ1(φ1,ψ1)(t)Φ1(φ2,ψ2)(t)|t0(ts)ν1Γ(ν)|f(s,φ1(s),ψ1(s))f(s,φ2(s),ψ2(s))|ds+12{1|Λ2|T0(s0(sx)ν1Γ(ν)|f(x,φ1(x),ψ1(x))f(x,φ2(x),ψ2(x))|dx+s0(sx)ρ1Γ(ρ)|g(x,φ1(x),ψ1(x))g(x,φ2(x),ψ2(x))|dx)ds+P2|Λ1|(T0(Ts)ν1Γ(ν)|f(s,φ1(s),ψ1(s))f(s,φ2(s),ψ2(s))|ds+T0(Ts)ρ1Γ(ρ)|g(s,φ1(s),ψ1(s))g(s,φ2(s),ψ2(s))|ds)+1|Λ2|ξη(s0(sx)ν1Γ(ν)|f(x,φ1(x),ψ1(x))f(x,φ2(x),ψ2(x))|dx+s0(sx)ρ1Γ(ρ)|g(x,φ1(x),ψ1(x))g(x,φ2(x),ψ2(x))|dx)ds+1|Λ1|mi=1ai(σi0(σis)ν1Γ(ν)|f(s,φ1(s),ψ1(s))f(s,φ2(s),ψ2(s))|ds+σi0(σis)ρ1Γ(ρ)|g(s,φ1(s),ψ1(s))g(s,φ2(s),ψ2(s))|ds)}{α(TνΓ(ν+1)+ϱ1)+βϱ2}(φ1φ2+ψ1ψ2),

    and

    |Φ2(φ1,ψ1)(t)Φ2(φ2,ψ2)(t)|t0(ts)ρ1Γ(ρ)|g(s,φ1(s),ψ1(s))g(s,φ2(s),ψ2(s))|ds+12{1|Λ2|T0(s0(sx)ν1Γ(ν)|f(x,φ1(x),ψ1(x))f(x,φ2(x),ψ2(x))|dx+s0(sx)ρ1Γ(ρ)|g(x,φ1(x),ψ1(x))g(x,φ2(x),ψ2(x))|dx)ds+P2|Λ1|(T0(Ts)ν1Γ(ν)|f(s,φ1(s),ψ1(s))f(s,φ2(s),ψ2(s))|ds+T0(Ts)ρ1Γ(ρ)|g(s,φ1(s),ψ1(s))g(s,φ2(s),ψ2(s))|ds)+1|Λ2|ξη(s0(sx)ν1Γ(ν)|f(x,φ1(x),ψ1(x))f(x,φ2(x),ψ2(x))|dx+s0(sx)ρ1Γ(ρ)|g(x,φ1(x),ψ1(x))g(x,φ2(x),ψ2(x))|dx)ds+1|Λ1|mi=1ai(σi0(σis)ν1Γ(ν)|f(s,φ1(s),ψ1(s))f(s,φ2(s),ψ2(s))|ds+σi0(σis)ρ1Γ(ρ)|g(s,φ1(s),ψ1(s))g(s,φ2(s),ψ2(s))|ds)}{αϱ1+β(TρΓ(ρ+1)+ϱ2)}(φ1φ2+ψ1ψ2).

    In view of the foregoing inequalities, it follows that

    Φ(φ1,ψ1)Φ(φ2,ψ2)=Φ1(φ1,ψ1)Φ1(φ2,ψ2)+Φ2(φ1,ψ1)Φ2(φ2,ψ2){α(TνΓ(ν+1)+2ϱ1)+β(TρΓ(ρ+1)+2ϱ2)}(φ1φ2,ψ1ψ2).

    Using the condition (3.9), we deduce from the above inequality that Φ is a contraction mapping. Consequently Φ has a unique fixed point by the application of contraction mapping principle. Hence there exists a unique solution for the problem (1.1) on J. The proof is finished.

    Example 3.1. Consider the following problem

    {CD1/2φ(t)=f(t,φ(t),ψ(t)),tJ:=[0,2],CD4/5ψ(t)=g(t,φ(t),ψ(t)),tJ:=[0,2],(φ+ψ)(0)+5/2(φ+ψ)(2)=1/2(φ+ψ)(1/4)+3/2(φ+ψ)(1/2),20(φψ)(s)ds3/42/3(φψ)(s)ds=1, (3.10)

    where ν=1/2,ρ=4/5,η=2/3,ζ=3/4,a1=1/2,a2=3/2,P1=1,P2=5/2,σ1=1/4,σ2=1/2,A=1,T=2, and f(t,φ,ψ) and g(t,φ,ψ) will be fixed later.

    Using the given data, we find that Λ1=1.5,Λ2=1.91666667, ϱ1=2.110627579,ϱ2=2.494392906, where Λ1,Λ2,ϱ1 and ϱ2 are respectively given by (2.5), (2.6), (3.4) and (3.5). For illustrating theorem 3.1, we take

    f(t,φ,ψ)=et516+t2(tan1φ+ψ+cost)andg(t,φ,ψ)=1(t+2)6(|φ|1+|ψ|+tψ+et). (3.11)

    Clearly f and g are continuous and satisfy the condition (H1) with μ1(t)=etcost516+t2,μ2(t)=et516+t2,μ3(t)=et1016+t2,κ1(t)=et(t+2)6,κ2(t)=1(t+2)6, and κ3(t)=12(t+2)6. Also

    μ2(2ϱ1+TνΓ(ν+1))+κ2(2ϱ2+TρΓ(ρ+1))0.398009902,

    and

    μ3(2ϱ1+TνΓ(ν+1))+κ3(2ϱ2+TρΓ(ρ+1))0.199004951<1.

    Thus all the conditions of theorem 3.1 hold true and hence the problem (3.10) with f(t,φ,ψ) and g(t,φ,ψ) given by (3.11) has at least one solution on [0,2].

    Next we demonstrate the application of Theorem 3.2. Let us choose

    f(t,φ,ψ)=ettan1φ+cosψ516+t2andg(t,φ,ψ)=1(2+t)6(|φ|2+|φ|+sinψ). (3.12)

    It is easy to show that the condition (H2) is satisfied with α1=α2=1/20=α and β1=1/64,β2=1/128 and so, β=1/64. Also α(TνΓ(ν+1)+2ϱ1)+β(TρΓ(ρ+1)+2ϱ2)0.39800990<1. Thus the hypothesis of Theorem 3.2 holds and hence its conclusion implies that the problem (3.10) with f(t,φ,ψ) and g(t,φ,ψ) given by (3.12) has a unique solution on [0,2].

    In this section, we consider a variant of the problem (1.1) involving a nonlinear Riemann-Liouville integral term in the last boundary condition given by

    {CDνφ(t)=f(t,φ(t),ψ(t)),tJ:=[0,T],CDρψ(t)=g(t,φ(t),ψ(t)),tJ:=[0,T],P1(φ+ψ)(0)+P2(φ+ψ)(T)=mi=1ai(φ+ψ)(σi),T0(φψ)(s)dsζη(φψ)(s)ds=1Γ(δ)T0(Ts)δ1h(s,φ(s),ψ(s))ds,δ>0. (4.1)

    Now we state a uniqueness result for the problem (4.1). We do not provide the proof of this result as it is similar to that of Theorem 3.2.

    Theorem 4.1. Let f,g,h:[0,1]×R2R be continuous functions and the following assumption holds:

    (¯H2) There exist positive constants αi,βi,γi,i=1,2, such that

    |f(t,φ1,ψ1)f(t,φ2,ψ2)|α1|φ1φ2|+α2|ψ1ψ2|,tJ,φi,ψiR,i=1,2;
    |g(t,φ1,ψ1)g(t,φ2,ψ2)|β1|φ1φ2|+β2|ψ1ψ2|,tJ,φi,ψiR,i=1,2;
    |h(t,φ1,ψ1)h(t,φ2,ψ2)|γ1|φ1φ2|+γ2|ψ1ψ2|,tJ,φi,ψiR,i=1,2.

    Then the problem (4.1) has a unique solution on J if

    γTδ|Λ2|Γ(δ+1)+α(TνΓ(ν+1)+2ϱ1)+β(TρΓ(ρ+1)+2ϱ2)<1, (4.2)

    where α=max{α1,α2},β=max{β1,β2},γ=max{γ1,γ2}, and ϱi,i=1,2 are defined in (3.4)-(3.5).

    Example 4.1. Let us consider the data given in Example 3.1 for the problem (4.1) with (3.12), h(t,φ,ψ)=(sinφ+cosψ+1/2)/t2+49 and δ=3/2. Then γ=1/7 and

    γTδ|Λ2|Γ(δ+1)+α(TνΓ(ν+1)+2ϱ1)+β(TρΓ(ρ+1)+2ϱ2)0.5565956<1.

    Clearly the assumptions of Theorem 4.1 are satisfied. Hence, by the conclusion of Theorem 4.1, the problem (4.1) with the given data has a unique solution on [0,2].

    We have studied a coupled system of nonlinear Caputo fractional differential equations supplemented with a new class of nonlocal multipoint-integral boundary conditions with respect to the sum and difference of the governing functions by applying the standard fixed point theorems. The existence and uniqueness results presented in this paper are not only new in the given configuration but also provide certain new results by fixing the parameters involved in the given problem. For example, our results correspond to the ones with initial-multipoint-integral and terminal-multipoint-integral boundary conditions by fixing P2=0 and P1=0 respectively in the present results. By taking A=0 in the present study, we obtain the results for the given coupled system of fractional differential equations with the boundary conditions of the form:

    P1(φ+ψ)(0)+P2(φ+ψ)(T)=mi=1ai(φ+ψ)(σi),T0(φψ)(s)ds=ζη(φψ)(s)ds,

    where the second (integral) condition means that the contribution of the difference of the unknown functions (φψ) on the domain (0,T) is equal to that on the sub-domain (η,ζ). Such a situation arises in heat conduction problems with sink and source. In the last section, we discussed the uniqueness of solutions for a variant of the problem (1.1) involving nonlinear Riemann-Liouville integral term in the last boundary condition of (1.1). This consideration further enhances the scope of the problem at hand. As a special case, the uniqueness result (Theorem 4.1) for the problem (4.1) corresponds to nonlinear integral boundary conditions for δ=1.

    This project was funded by the Deanship of Scientific Research (DSR), King Abdulaziz University, Jeddah, Saudi Arabia under grant no. (KEP-PhD-80-130-42). The authors, therefore, acknowledge with thanks DSR technical and financial support. The authors also thank the reviewers for their useful remarks that led to the improvement of the original manuscript.

    The authors declare that they have no competing interests.



    [1] Z. Lin, An empirical investigation of user and system recommendations in e-commerce, Decis. Support Syst., 68 (2014), 111–124. https://doi.org/10.1016/j.dss.2014.10.003 doi: 10.1016/j.dss.2014.10.003
    [2] T. Iba, K. Nemoto, B. Peters, P. A. Gloor, Analyzing the creative editing behavior of wikipedia editors: Through dynamic social network analysis, Proc.-Soc. Behav. Sci., 2 (2010), 6441–6456. https://doi.org/10.1016/j.sbspro.2010.04.054 doi: 10.1016/j.sbspro.2010.04.054
    [3] T. R. Liyanagunawardena, A. A. Adams, S. A. Williams, MOOCs: A systematic study of the published literature 2008–2012, Int. Rev. Res. Open Distrib. Learn., 14 (2013), 202–227. https://doi.org/10.19173/irrodl.v14i3.1455 doi: 10.19173/irrodl.v14i3.1455
    [4] Q. Liu, S. Wu, L. Wang, T. Tan, Predicting the next location: A recurrent model with spatial and temporal contexts, in Thirtieth AAAI Conference on Artificial Intelligence, 30 (2016). https://doi.org/10.1609/aaai.v30i1.9971
    [5] S. Wu, Q. Liu, P. Bai, L. Wang, T. Tan, SAPE: A system for situation-aware public security evaluation, in Thirtieth AAAI Conference on Artificial Intelligence, 30 (2016). https://doi.org/10.1609/aaai.v30i1.9828
    [6] Z. Hou, X. Lv, Y. Zhou, L. Bu, Q. Ma, Y. Wang, A dynamic graph Hawkes process based on linear complexity self-attention for dynamic recommender systems, PeerJ. Comput. Sci., 9 (2023), 1368. https://doi.org/10.7717/peerj-cs.1368 doi: 10.7717/peerj-cs.1368
    [7] Y. Zheng, X. Yi, M. Li, R. Li, Z. Shan, E. Chang, et al., Forecasting fine-grained air quality based on big data, in 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, (2015), 2267–2276. https://doi.org/10.1145/2783258.2788573
    [8] X. Wang, X. He, M. Wang, F. Feng, T. Chua, Neural graph collaborative filtering, in 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval, (2019), 165–174. https://doi.org/10.1145/3331184.3331267
    [9] X. He, K. Deng, X. Wang, Yan Li, Y. Zhang, M. Wang, LightGCN: Simplifying and powering graph convolution network for recommendation, in 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, (2020), 639–648. https://doi.org/10.1145/3397271.3401063
    [10] Y. Koren, R. Bell, C. Volinsky, Matrix factorization techniques for recommender systems, Computer, 42 (2009), 30–37. https://doi.org/10.1109/MC.2009.263 doi: 10.1109/MC.2009.263
    [11] D. Lee, H. S. Seung, Algorithms for non-negative matrix factorization, in 13th International Conference on Neural Information Processing Systems, (2000), 535–541.
    [12] R. Salakhutdinov, A. Mnih, Probabilistic matrix factorization, in 20th International Conference on Neural Information Processing Systems, (2007), 1257–1264.
    [13] S. Raza, C. Ding, Progress in context-aware recommender systems-An overview, Comput. Sci. Rev., 31 (2019), 84–97. https://doi.org/10.1016/j.cosrev.2019.01.001 doi: 10.1016/j.cosrev.2019.01.001
    [14] G. Adomavicius, K. Bauman, A. Tuzhilin, M. Unger, Context-aware recommender systems: From foundations to recent developments, in Recommender Systems Handbook, (2022), 211–250. https://doi.org/10.1007/978-1-0716-2197-4_6
    [15] C. Wu, A. Ahmed, A. Beutel, A. J. Smola, H. Jing, Recurrent recommender networks, in Tenth ACM International Conference on Web Search and Data Mining, (2017), 495–503. https://doi.org/10.1145/3018661.3018689
    [16] S. Kumar, X. Zhang, J. Leskovec, Predicting dynamic embedding trajectory in temporal interaction networks, in 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, (2019), 1269–1278. https://doi.org/10.1145/3292500.3330895
    [17] X. Li, M. Zhang, S. Wu, Z. Liu, L. Wang, P. S. Yu, Dynamic graph collaborative filtering, in 2020 IEEE International Conference on Data Mining, (2020), 322–331. https://doi.org/10.1109/ICDM50108.2020.00041
    [18] Z. Kefato, S. Girdzijauskas, N. Sheikh, A. Montresor, Dynamic embeddings for interaction prediction, in Web Conference 2021, (2021), 1609–1618. https://doi.org/10.1145/3442381.3450020
    [19] S. Wu, F. Sun, W. Zhang, X. Xie, B. Cui, Graph neural networks in recommender systems: A survey, ACM Comput. Surv., 55 (2022), 1–37. https://doi.org/10.1145/3535101 doi: 10.1145/3535101
    [20] P. Covington, J. Adams, E. Sargin, Deep neural networks for YouTube recommendations, in 10th ACM Conference on Recommender Systems, (2016), 191–198. https://doi.org/10.1145/2959100.2959190
    [21] H. Dai, Y. Wang, R. Trivedi, L. Song, Recurrent coevolutionary latent feature processes for continuous-time recommendation, in 1st Workshop on Deep Learning for Recommender Systems, (2016), 29–34. https://doi.org/10.1145/2988450.2988451
    [22] Y. K. Tan, X. Xu, Y. Liu, Improved recurrent neural networks for session-based recommendations, in 1st Workshop on Deep Learning for Recommender Systems, (2016), 17–22. https://doi.org/10.1145/2988450.2988452
    [23] Z. Wang, W. Wei, G. Cong, X. Li, X. Mao, M. Qiu, Global context enhanced graph neural networks for session-based recommendation, in 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, (2020), 169–178. https://doi.org/10.1145/3397271.3401142
    [24] S. Wu, Y. Tang, Y. Zhu, L. Wang, X. Xie, T. Tan, Session-based recommendation with graph neural networks, in AAAI Technical Track: AI and the Web, 33 (2019), 346–353. https://doi.org/10.1609/aaai.v33i01.3301346
    [25] C. Xu, P. Zhao, Y. Liu, V. S. Sheng, J. Xu, F. Zhuang, Graph contextualized self-attention network for session-based recommendation, in Twenty-Eighth International Joint Conference on Artificial Intelligence, (2019), 3940–3946. https://doi.org/10.24963/ijcai.2019/547
    [26] B. Hidasi, A. Karatzoglou, L. Baltrunas, D. Tikk, Session-based recommendations with recurrent neural networks, preprint, arXiv: 1511.06939.
    [27] J. Li, P. Ren, Z. Chen, Z. Ren, T. Lian, J. Ma, Neural attentive session-based recommendation, in 2017 ACM on Conference on Information and Knowledge Management, (2017), 1419–1428. https://doi.org/10.1145/3132847.3132926
    [28] Q. Liu, S. Wu, D. Wang, Z. Li, L. Wang, Context-aware sequential recommendation, in 2016 IEEE 16th International Conference on Data Mining (ICDM), (2016), 1053–1058. https://doi.org/10.1109/ICDM.2016.0135
    [29] J. You, Y. Wang, A. Pal, P. Eksombatchai, C. Rosenburg, Hierarchical temporal convolutional networks for dynamic recommender systems, in The World Wide Web Conference, (2019), 2236–2246. https://doi.org/10.1145/3308558.3313747
    [30] Y. Wang, N. Du, R. Trivedi, L. Song, Coevolutionary latent feature processes for continuous-time user-item interactions, in 30th Conference on Neural Information Processing Systems, (2016), 1–9.
    [31] Q. Wu, Y. Gao, X. Gao, P. Weng, G. Chen, Dual sequential prediction models linking sequential recommendation and information dissemination, in 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, (2019), 447–457. https://doi.org/10.1145/3292500.3330959
    [32] R. Trivedi, M. Farajtabar, P. Biswal, H. Zha, DyRep: Learning representations over dynamic graphs, in International Conference on Learning Representations 2019 Conference, (2019).
    [33] A. Beutel, P. Covington, S. Jain, C. Xu, J. Li, V. Gatto, et al., Latent cross: Making use of context in recurrent recommender systems, in Eleventh ACM International Conference on Web Search and Data Mining, (2018), 46–54. https://doi.org/10.1145/3159652.3159727
    [34] Y. Zhu, H. Li, Y. Liao, B. Wang, Z. Guan, H. Liu, et al., What to do next: Modeling user behaviors by time-LSTM, in 26th International Joint Conference on Artificial Intelligence, (2017), 3602–3608.
    [35] Y. Zhang, X. Yang, J. Ivy, M. Chi, ATTAIN: Attention-based time-aware LSTM networks for disease progression modeling, in Twenty-Eighth International Joint Conference on Artificial Intelligence, (2019), 4369–4375. https://doi.org/10.24963/ijcai.2019/607
    [36] W. Kang, J. McAuley, Self-attentive sequential recommendation, in 2018 IEEE International Conference on Data Mining, (2018), 197–206. https://doi.org/10.1109/ICDM.2018.00035
    [37] D. Hendrycks, K. Gimpel, Gaussian error linear units (GeLUs), preprint, arXiv: 1606.08415.
    [38] M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, L. Chen, MobileNetV2: Inverted residuals and linear bottlenecks, in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, (2018), 4510–4520. https://doi.org/10.1109/CVPR.2018.00474
    [39] Y. N. Dauphin, A. Fan, M. Auli, D. Grangier, Language modeling with gated convolutional networks, in 34th International Conference on Machine Learning, 70 (2017), 933–941.
    [40] R. K. Srivastava, K. Greff, J. Schmidhuber, Highway networks, preprint, arXiv: 1505.00387.
    [41] S. Hochreiter, J. Schmidhuber, Long Short-Term Memory, Neural Comput., 9 (1997), 1735–1780. https://doi.org/10.1162/neco.1997.9.8.1735 doi: 10.1162/neco.1997.9.8.1735
    [42] J. Hu, L. Shen, G. Sun, Squeeze-and-excitation networks, in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, (2018), 7132–7141. https://doi.org/10.1109/CVPR.2018.00745
    [43] C. Santos, M. Tan, B. Xiang, B. Zhou, Attentive pooling networks, preprint, arXiv: 1602.03609.
    [44] R. Hadsell, S. Chopra, Y. LeCun, Dimensionality reduction by learning an invariant mapping, in 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'06), (2006), 1735–1742. https://doi.org/10.1109/CVPR.2006.100
    [45] S. Yang, W. Yu, Y. Zheng, H. Yao, T. Mei, Adaptive semantic-visual tree for hierarchical embeddings, in 27th ACM International Conference on Multimedia, (2019), 2097–2105. https://doi.org/10.1145/3343031.3350995
    [46] X. He, K. Deng, X. Wang, Y. Li, Y. Zhang, M. Wang, Lightgcn: Simplifying and powering graph convolution network for recommendation, in 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, (2020), 639–648. https://doi.org/10.1145/3397271.3401063
    [47] C. Hsieh, L. Yang, Y. Cui, T. Lin, S. Belongie, D. Estrin, Collaborative metric learning, in 26th International Conference on World Wide Web, (2017), 193–201. https://doi.org/10.1145/3038912.3052639
    [48] B. Fu, W. Zhang, G. Hu, X. Dai, S. Huang, J. Chen, Dual side deep context-aware modulation for social recommendation, in Web Conference 2021, (2021), 2524–2534. https://doi.org/10.1145/3442381.3449940
    [49] T. Mikolov, I. Sutskever, K. Chen, G. Corrado, J. Dean, Distributed representations of words and phrases and their compositionality, in 26th International Conference on Neural Information Processing Systems, 2 (2013), 3111–3119.
    [50] R. Girshick, J. Donahue, T. Darrell, J. Malik, Rich feature hierarchies for accurate object detection and semantic segmentation, in 2014 IEEE Conference on Computer Vision and Pattern Recognition, (2014), 580–587. https://doi.org/10.1109/CVPR.2014.81
    [51] N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, R. Salakhutdinov, Dropout: A simple way to prevent neural networks from overfitting, J. Mach. Learn. Res., 15 (2014), 1929–1958.
    [52] B. Hidasi, D. Tikk, Fast ALS-based tensor factorization for context-aware recommendation from implicit feedback, in 2012th European Conference on Machine Learning and Knowledge Discovery in Databases, (2012), 67–82.
    [53] X. Glorot, Y. Bengio, Understanding the difficulty of training deep feedforward neural networks, in 13th International Conference on Artificial Intelligence and Statistics (AISTATS) 2010, (2010), 249–256.
    [54] J. Chung, C. Gulcehre, K. Cho, Y. Bengio, Empirical evaluation of gated recurrent neural networks on sequence modeling, preprint, arXiv: 1412.3555.
    [55] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, et al., Attention is all you need, in 31st International Conference on Neural Information Processing Systems, (2017), 6000–6010.
    [56] T. Silveira, M. Zhang, X. Lin, Y. Liu, S. Ma, How good your recommender system is? A survey on evaluations in recommendation, Int. J. Mach. Learn. Cybern., 10 (2019), 813–831. https://doi.org/10.1007/s13042-017-0762-9 doi: 10.1007/s13042-017-0762-9
  • This article has been cited by:

    1. Haitham Qawaqneh, Hasanen A. Hammad, Hassen Aydi, Exploring new geometric contraction mappings and their applications in fractional metric spaces, 2024, 9, 2473-6988, 521, 10.3934/math.2024028
  • Reader Comments
  • © 2024 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(1818) PDF downloads(87) Cited by(3)

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog