Recommendation algorithm based on weighted trust relationship enhancement

Abstract

The invention discloses a recommendation algorithm based on weighted trust relationship enhancement (TECosRA), which comprises the following steps of: firstly, carrying out resource initialization on an item selected by a user, then, calculating a resource value obtained by the user from the item by using CosRA similarity in a resource diffusion process, then, introducing a trust relationship of the user, enabling the user to obtain a trust relationship resource value, and carrying out resource allocation on the trust relationship resource value. And linearly combining the trust resource value of the control user of the adjustable parameter with the project resource value of the user, carrying out descending sorting on the obtained final resource values of the articles, excluding the articles selected by the user, and generating a recommendation list with the length of L. According to the algorithm, the accuracy and diversity of a recommendation system are improved, a certain balance is achieved in accuracy and diversity, the positive effect of a trust relationship in a social network on the recommendation algorithm is enhanced, and the negative influence of data sparsity and cold start problems on the recommendation algorithm is relieved to a certain extent.

Description

Technical Field

[0001] The present invention relates to the technical field of recommendation algorithms, and more specifically, to a recommendation algorithm based on weighted trust relationship enhancement (TECosRA). Background Art

[0002] In social networks, according to Nielsen research, people are more inclined to trust recommendations from friends than from strangers. The popularity of social networks has enabled people to become friends and trust each other online, giving rise to community-based recommendation systems that are receiving increasing attention. This type of recommendation system is called a social recommendation system, which makes predictions based on user profiles and relationships between users. Studies have shown that users trust recommendations from friends more than recommendation systems, and that recommendations based on social influence are more important than recommendations based on similarities in user historical behavior. For example, by analyzing social relationships, the accuracy of online shopping recommendations can be improved.

[0003] There are many studies on recommendation algorithms based on social networks, where social relationships are very complex and diverse. The basic idea of ​​collaborative filtering recommendation algorithms is to use the similarity between users or projects to make recommendations or predictions. It is to calculate the similarity between users of a specified project, find a group of users with the same preferences, and then analyze the common preferences of users to recommend target users and finally make recommendations to generate a recommendation list. The advantage of collaborative filtering algorithms is that they do not focus on the content of the project itself, but mainly recommend resources based on user or project similarity. The system only needs to obtain enough project evaluations to reliably recommend projects. However, this method needs to be evaluated based on a large amount of data, and needs to rely on the accuracy of similarity to improve the accuracy of recommendations.

[0004] CN117807329A discloses a collaborative filtering recommendation method based on rating weighting under a trust relationship, including: constructing a user-project category matrix by combining user-project rating items and project category data sets, and then obtaining a user-project category matrix under a trust relationship based on the trust relationship; clustering the user category matrix under the trust relationship, and assigning users of the same category to a cluster, thereby obtaining k clusters; calculating the similarity between the target user and other users in the cluster, and finding M users with the highest similarity to form a neighbor set of the target user; applying the improved weighted Slope One algorithm to the neighbor set of the target user to predict the rating of its unrated items, and exporting the top N ranking results based on the predicted rating to the recommendation list. This method improves the performance of the recommendation algorithm and the accuracy of the recommendation by increasing the trust links between users. However, the patent still needs to ensure the accuracy of the recommendation algorithm based on the situation of a large number of users and sufficient data. The data sparsity of the trust relationship in the social network will reduce the performance of the recommendation algorithm, causing users to fall into the dilemma of "information cocoon". Summary of the invention

[0005] The main technical problem to be solved by the present invention is to provide a recommendation algorithm (TECosRA) based on weighted trust relationship enhancement, in order to address the shortcomings of the prior art that the data sparsity of trust relationships in social networks will reduce the performance of the recommendation algorithm, causing users to fall into the dilemma of "information cocoons".

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A recommendation algorithm based on weighted trust relationship enhancement, comprising the following steps:

[0008] S1. For target user u i The selected project α performs resource initialization;

[0009] S2. Use CosRA similarity to calculate the target user u i The resource value f obtained from the project ′(i) ;

[0010] S3. Introducing target user u i Trust relationship, get the target user u i Trust relationship resource value The resource value The target user u i Other trusted users j About Project α The average resource value is expressed as:

[0011]

[0012] in represents the target user u i Trusted User j For items α The average rating value, k i Represents user u i The degree, f′ (i) Represented as target user u i The resource value of all items, m represents the total number of users in the network;

[0013] S4. Control the target user u with adjustable parameters i The final resource allocation is expressed as:

[0014]

[0015] Among them, λ is an adjustable parameter, b ij Represents user u i For user u j Trust?

[0016] S5. Arrange the resource values ​​obtained by the items in descending order and exclude the selected items to obtain a recommendation list with a length of L.

[0017] Furthermore, resource initialization is expressed as:

[0018]

[0019] Among them, the target user u i No item selected α Then a iα =0, target user u i Selected item α Then a iα =1. Furthermore, the target user u i Resource value of all items f ′(i) It is expressed as:

[0020] f ′(i) =S CosRA f (i)

[0021] Where S CosRA is the CosRA similarity.

[0022] Furthermore, the S CosRA It is expressed as:

[0023]

[0024] Among them, α and β represent item o respectively. α and items β , aiα and a iβ Represents user u i For items α and items β The rating, k α , k β and k i Represents items o α , o β and user u i The degree of η1 and η2 are expressed as values ​​in the range [-1,1].

[0025] Furthermore, the values ​​of η1 and η2 with the best recommendation performance of the CosRA algorithm are -0.5.

[0026] Furthermore, the value range of λ is 0.275 to 0.575. Preferably, in the FriendFeed dataset, λ is 0.375; in the Epinions dataset, λ is 0.275 or 0.575.

[0027] Furthermore, if the target user u i Trust user u j , then b ij =1, otherwise b ij =0; when b ij = 0, the target user u i The final resource allocation obtained is expressed as:

[0028] f ′(i) =S CosRA f (i) .

[0029] Furthermore, the evaluation indicators of the accuracy of the algorithm include AUC, Ranking Score, nDCG, precision, recall and F1 index; the diversity indicators include one or more of Hamming distance, internal similarity and novelty.

[0030] Furthermore, the length of the recommendation list does not exceed 50.

[0031] Compared with the prior art, the beneficial effects are:

[0032] The recommendation algorithm based on weighted trust relationship enhancement of the present invention first initializes the resources of the items selected by the user, then uses CosRA similarity to calculate the resource value obtained by the user from the item, then introduces the user's trust relationship, and performs a weighted effect on the user's resources based on the score similarity obtained by social similarity, and uses adjustable parameters to control the final allocation of the user's resources. Finally, the final resource values ​​obtained by the items are sorted in descending order to generate a recommendation list with a length of L.

[0033] The CosRA similarity index adopted by the present invention not only utilizes the nodes with large degrees to expand the scope, but also increases the weight of the nodes with small degrees, which can balance the dilemma of accuracy and diversity of the recommendation algorithm. In addition, the trust relationship in the user's social network is introduced into the algorithm. On the one hand, the data of the "user-item" rating is supplemented. The recommendation algorithm calculates the taste similarity between users based on the user's historical behavior and recommends items selected by similar users. After introducing the trust relationship, the present invention uses social relationships to improve the user's behavior pattern and expands the scope that can be used to calculate user similarity, thereby alleviating the cold start problem. On the other hand, the resource value of the item is weighted by the user's trust relationship, strengthening the positive impact of the trust relationship on the user's behavior and improving the accuracy of the recommendation algorithm. At the same time, the source of the user's recommended items includes similar users and trusted users, which expands the scope of recommended items and plays a positive role in breaking the "information cocoon" of the recommendation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is the influence diagram of the adjustable parameters.

[0035] Figure 2 is a graph of optimal values ​​of adjustable parameters.

[0036] Figure 3 This is a graph showing how the accuracy of the recommendation results is affected by the length of the list.

[0037] Figure 4 This is a graph showing how the diversity and novelty of recommendation results are affected by the length of the list. DETAILED DESCRIPTION

[0038] The present invention will be further explained and illustrated below in conjunction with the embodiments, but the specific embodiments do not limit the present invention in any form.

[0039] Example 1

[0040] This embodiment provides a recommendation algorithm based on weighted trust relationship enhancement. i 、u j Indicates that items are marked with o α , o β Indicates that the steps include:

[0041] S1. For target user u i The selected project α performs resource initialization, which is expressed as:

[0042]

[0043] Among them, the target user u i No item selected α Then a iα =0, otherwise a iα=1.

[0044] S2. Use CosRA similarity to calculate the resource value f′ obtained by the target user from the project (i) :

[0045] f ′(i) =S CosRS f (i)

[0046] Where S CosRA is the CosRA similarity, expressed as:

[0047]

[0048] Among them, α and β represent item o respectively. α and items β , a iα and a iβ Represents user u i For items α and items β The rating, k a , k β and k i Represents items o α , o β and user u i The degree of η1 and η2 are expressed as values ​​in the range [-1,1], with an interval of 0.05. The optimal parameter value is η1=η2=-0.5.

[0049] S3. Introduce the trust relationship of the target user and obtain the trust relationship resource value of the target user The resource value is the target user u i Other trusted users j For project o α The average resource value is expressed as:

[0050]

[0051] in Represents user u i Trusted User j For items α The average rating value, k i Represents user u i The degree, f′ (i) It is represented as the resource value of all projects of user i, and m represents the total number of users in the network.

[0052] S4. Control the final allocation of user resources through adjustable parameters, expressed as:

[0053]

[0054] Among them, λ is an adjustable parameter, b ij Represents user u i For user u j Trust or not, if user u i Trust user u j , then b ij =1, otherwise b ij =0; when b ij = 0, the target user u i The final resource allocation obtained is expressed as:

[0055] f ′(i) =S CosRA f (i)

[0056] S5. Arrange the resource values ​​obtained by the items in descending order to obtain a recommendation list of length L.

[0057] Example 2

[0058] This embodiment provides an experimental example of the influence of adjustable parameter scaling. This embodiment is based on 8 evaluation indicators: accuracy indicators: AUC, Ranking Score (RS), precision (P), recall (R) and F1; diversity indicators: Hamming distance (H), internal similarity (I) and novelty (N). The specifics are as follows:

[0059] (1) The AUC indicator is used to test the accuracy of the recommendation algorithm in predicting items in a global context. For an algorithm intervention, the AUC value of a randomly sorted recommendation list is around 0.5, with a maximum value of 1. The larger the AUC value, the higher the prediction accuracy of the algorithm, so the AUC value of the algorithm is proportional to its recommendation accuracy. Its formula is expressed as:

[0060]

[0061] Where m represents the total number of users, N represents the number of independent comparisons, N1 represents the number of times the objects collected by the target user have more resources than the uncollected objects, and N2 represents the number of times the objects collected by the target user have the same resources as the uncollected objects.

[0062] (2) Accuracy (P) is used to test the proportion of correct predictions in the target user's recommendation list. When no items in the target user's recommendation list are predicted, P = 0, and when all predictions are correct, P = 1. Therefore, the higher the accuracy, the more the algorithm's prediction fits the user's actual situation. Its formula is:

[0063]

[0064] where d i(L) is the number of objects in common between the test set and the recommendation list of length L.

[0065] (3) Recall (R) is used to test the proportion of items in the target user test set that are predicted correctly. When all items in the target user test set are predicted correctly, R = 0, and when all items in the target user test set are predicted correctly, R = 1. Therefore, the higher the recall, the more the prediction of the algorithm meets the user's interests. Its formula is expressed as:

[0066]

[0067] Where D i (L) is the number of objects in the test set.

[0068] (4) The F1 index is an indicator that comprehensively evaluates the performance of an algorithm by considering both accuracy and recall. The F1 value is proportional to the accuracy of the algorithm. Its formula is:

[0069]

[0070] Where P i (L) is the Precision value of user i, R i (L) is the Recall value of user i.

[0071] (5) RS is a ranking evaluation metric similar to AUC. RS is used to measure the relative ranking of items in the target user's test set in the user's recommendation list. The higher the position of the item predicted correctly by the algorithm in the recommendation list, the lower the RS value, indicating that the ranking accuracy of the item predicted by the algorithm is better. The sum of AUC and RS is approximately 1. Its formula is expressed as:

[0072]

[0073] Where |E p | represents the size of the test set, p α is the position of the object in the recommendation list, l i is the number of objects that user i has not collected in the training set.

[0074] (6) Internal similarity (I) is used to evaluate the diversity of items in the user recommendation list. The lower the I value, the more diverse the items recommended by the algorithm to the target user. Its formula is:

[0075]

[0076] in Represents the cosine similarity between the recommended objects α and β in the recommendation list of user i.

[0077] (7) Hamming distance (H) evaluates the degree of difference between items in recommendation lists of different users. When the items in the recommendation lists of two different users are consistent, H = 0, and when they are completely inconsistent, H = 1. This means that the larger the H value, the greater the difference in the user's recommendation list, and the more diverse the items recommended by the algorithm. Its formula is expressed as:

[0078]

[0079] Where C(i,j) represents the number of identical objects in the recommendation lists of length L of user i and user j.

[0080] (8) Novelty (N) is an indicator of the popularity of items in the user recommendation list. The larger the N value, the more popular the recommended item is. The smaller the N value, the less popular the recommended item is. The higher the novelty of the algorithm, the more suitable it is for personalized recommendation scenarios. The formula is:

[0081]

[0082] where k α is the degree of the recommended object α in the target user's recommendation list.

[0083] When the three accuracy indicators used in this embodiment are evaluated on the FriendFeed dataset and the Epinions dataset, there is a peak in accuracy as λ changes. Figure 1 As shown in Figure 1, on the FriendFeed dataset, the best accuracy result is when λ = 0.375. On the Epinions dataset, the global accuracy (AUC and RS) results are when λ = 0.575. After that, the accuracy of the algorithm begins to gradually decrease.

[0084] from Figure 2 From the trend of (f), the F1(L) value on the Epinions dataset is at its highest when λ is approximately equal to 0.275. When λ exceeds 0.275, the F1(L) value begins to decrease gradually, which is different from the performance on the FriendFeed dataset.

[0085] During the experiment, it was observed that there is a phenomenon in the Epinions dataset: when performing k-fold cross validation, the dataset needs to be divided into k parts. At this time, it is found that some items are not selected by any user in the training set, but are selected by users in the test set. Therefore, when using the training set for prediction, there is no historical record and it is impossible to predict such items. AUC and RS are evaluated under global conditions, while F1(L) is evaluated based on the length of the recommendation list L. Therefore, the values ​​of AUC, RS, and F1(L) on the Epinions dataset are at the optimal lambda. * There is some difference.

[0086] In order to determine the optimal value of the adjustable parameter λ, three accuracy indicators that depend on the length of the recommendation list L, namely, precision P, recall R, and F1 value, are used for testing, as shown in Figure 2 The (ab,de) shows that the optimal parameter λ * It is not very sensitive to the change of the length L of the recommendation list. For the P and R indicators, the best parameter is 0.575 on the FriendFeed dataset, as shown in Figure 2 (a). The best parameter is 0.275 on the Epinions dataset. Figure 2 (d). The same trend is also seen for the F1 indicator ( Figure 2 (b) and Figure 2 (b)). In addition, Figure 2 In (a) and 2(d), the optimal parameter values ​​λ of the accuracy indicators P, R and F1 are determined by experiments. * Results on FriendFeed and Epinions. The results also confirm that the accuracy indicators (AUC and RS) that do not depend on L are FriendFeed (λ * )≈0.375, Epinions(λ * )≈0.275. Figure 2 It can be observed that the optimal parameter values ​​for the five accuracy indicators on the FriendFeed dataset are the same. On the Epinions dataset, the λ corresponding to the best values ​​of AUC and RS are consistent, and the λ corresponding to the best values ​​of P(L), R(L), and F1(L) are also consistent, indicating that λ is stable. This shows the universality of the TECosRA algorithm under the optimal parameters, and that trust relationships can indeed improve the performance of recommendation algorithms.

[0087] Example 3

[0088] This example compares the performance of the method of the present invention with that of five benchmark algorithms (MD, HC, UCF, CosRA, and CosRA+T) on the same dataset, using nine accuracy evaluation metrics, including two L-independent metrics (AUC and RS) and three L-dependent metrics (nDCG, P, R, and F1), two L-dependent diversity metrics (H and I), and one L-dependent novelty metric N. In the experiment, the length of the recommendation list was first set to L = 10. In the comparison with the benchmark method, the TECosRA algorithm used the optimal parameter value λ for each dataset in all evaluation metrics. * (FriendFeed * =0.375, take λ on Epinions * =0.275), and the calculation result is the average value after 1000 independent calculations. The results of the recommended algorithm are shown in Table 1:

[0089] Table 1

[0090] Epinions AUC RS P R F1 H I N MD 0.8255 0.1750 0.0189 0.0590 0.0286 0.6718 0.1134 229 HC 0.7795 0.2211 0.0051 0.0153 0.0077 0.9749 0.0243 5 UCF 0.8190 0.1921 0.0170 0.05371 0.02586 0.5673 0.1311 256 CosRA 0.8301 0.1703 0.0221 0.0629 0.0327 0.9520 0.0868 100 CosRA+T 0.8321 0.1682 0.0226 0.0651 0.0336 0.9582 0.0890 94 TECosRA 0.8449 0.1551 0.0234 0.0718 0.0357 0.8364 0.1112 170

[0091]

[0092]

[0093] It can be seen from Table 1 that the performance of the TECosRA algorithm of the present invention is better than all benchmark algorithms on the FriendFeed and Epinions datasets. The AUC values ​​of the algorithm are 0.8449 and 0.9151 on the FriendFeed and Epinions datasets, respectively. In terms of accuracy indicators, the performance of the algorithm is better than that of the benchmark algorithm. Specifically, the P, R, and F1 values ​​of the algorithm on the FriendFeed and Epinions datasets are the highest among all algorithms. In terms of the accuracy indicator RS based on sorting, the RS value of the algorithm is the lowest among all benchmark algorithms, indicating that the new algorithm has the best performance among all benchmark algorithms in terms of sorting accuracy. In terms of diversity indicators, the performance of the algorithm of the present invention is better than that of the UCF algorithm and is close to that of the MD algorithm. The I value of the algorithm of the present invention on the Epinions and FriendFeed datasets is much lower than that of the UCF algorithm, the H value is much higher than that of the UCF algorithm, and the N value is in the middle position among the benchmark algorithms. Therefore, the performance of the algorithm described in the present invention is better than several benchmark algorithms as a whole.

[0094] Example 4

[0095] This embodiment studies the influence of the length L of the recommendation list on the performance of the algorithm, conducts experiments by changing the length of the recommendation list, and analyzes the influence on the recommendation results. The length of the recommendation list L in the recommendation algorithm may affect the evaluation of the recommendation performance under the evaluation indicators related to it, including three accuracy indicators (P(L), R(L) and F1(L)), two diversity indicators (H(L) and I(L)) and one novelty indicator (N(L)).

[0096] In order to explore the impact of the recommendation list length L on the algorithm performance, the recommendation list length L is changed from 0 to 100 to explore how L affects the performance of the recommendation algorithm. The TECosRA algorithm is compared with five other benchmark methods.

[0097] Depend on Figure 3 The results of the three accuracy indicators of the recommendation algorithm shown in the figure on the FriendFeed and Epinions datasets as the length of the recommendation list L changes show that for all the benchmark methods selected in this embodiment, the performance increases as the length of the recommendation list L increases, and the value of P(L) decreases (see Figure 3 (a) and 3(d)), while the value of R(L) increases ( Figure 3 (b) and 3(e)). As the length of the recommendation list L increases, the F1 value first increases and then decreases, and F1(L) on FriendFeed and Epinions reaches its maximum value when L≈5. In addition, the performance of the TECosRA algorithm is relatively better than that of other benchmark methods under different recommendation list lengths L, and the effect is more obvious when L is near the optimal value of its adjustable parameters. It can also be observed from the figure that the accuracy index of the TECosRA algorithm of the present invention is significantly improved compared with the CosRA+T algorithm, the MD algorithm and the UCF algorithm compete with each other, and the HC algorithm has the lowest accuracy.

[0098] Depend on Figure 4 The results of the correlation between the two diversity indicators and the novelty indicator shown in FIG. 1 show that for the Hamming distance (H(L)), the value of H(L) decreases as L increases on the FriendFeed and Epinions datasets, as shown in Figure 4 (a) and Figure 4 (d). The HC algorithm performs best, followed by the CosRA algorithm and the CosRA+T algorithm. The performance of the TECosRA algorithm on H(L) is close to that of the MD algorithm, and the worst performance is the UCF algorithm. For the internal similarity I(L), as the length of the recommendation list L increases, the value of I(L) increases rapidly at first, and then decreases slowly (see Figure 4 (b) and Figure 4(e)), in this evaluation index, the maximum I(L) value of UCF indicates that it has the worst performance. The improved new algorithm is better than the UCF algorithm and close to the performance of the MD algorithm. The HC algorithm is better than all other methods on the dataset because it has the smallest I(L) value. In terms of novelty N(L), the N(L) value will decrease rapidly at the beginning as the length of the recommendation list L increases, and then show a slow decreasing trend, such as Figure 4 (c) and Figure 4 (f). The HC algorithm performs best here, and the new algorithm of the present invention is in the middle of all algorithms, indicating that the improved new algorithm has certain advantages in this indicator. From the results, it can be observed that, except for HC, the TECosRA algorithm outperforms most benchmark methods in the novelty indicator. After the recommendation list exceeds about 50, the algorithm's results begin to become gradually insensitive to the change of L, which shows that the weighted trust-enhanced recommendation algorithm has higher diversity and better novelty in the recommendation of uncollected objects than other benchmark methods.

[0099] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A recommendation algorithm based on weighted trust relationship enhancement, characterized in that the steps include: S1. For target user u i The selected project α performs resource initialization; S2. Use CosRA similarity to calculate the target user u i The resource value f obtained from the project ′(i) ; S3. Introducing target user u i Trust relationship, get the target user u i Trust relationship resource value The resource value The target user u i Other trusted users j For project o α The average resource value is expressed as: in represents the target user u i Trusted User j For items α The average rating value, k i Represents user u i The degree, f′ (i) Represented as target user u i The resource value of all items, m represents the total number of users in the network; S4. Control the target user u with adjustable parameters i The final resource allocation is expressed as: Among them, λ is an adjustable parameter, b ij Represents user u i For user u j Trust? S5. Arrange the resource values ​​obtained by the items in descending order and exclude the selected items to obtain a recommendation list with a length of L.

2. The recommendation algorithm based on weighted trust relationship enhancement according to claim 1, characterized in that: Resource initialization is represented as: Among them, the target user u i No item selected α Then a iα =0, target user u i Selected item α Then a iα =1.

3. The recommendation algorithm based on weighted trust relationship enhancement according to claim 1 is characterized in that: Target user i Resource value of all items f ′(i) It is expressed as: f ′(i) =S CosRA f (i) Where S CosRA is the CosRA similarity.

4. The recommendation algorithm based on weighted trust relationship enhancement according to claim 3 is characterized in that: The S CosRA It is expressed as: Among them, α and β represent item o respectively. α and items β , a iα and a iβ Represents user u i For items α and items β The rating, k α , k β and k i Represents items o α , o β and user u i The degree of η1 and η2 are expressed as values ​​in the range [-1,1].

5. The recommendation algorithm based on weighted trust relationship enhancement according to claim 4 is characterized in that: The η1 and v2 values ​​with the best recommendation performance of the CosRA algorithm are -0.

5.

6. The recommendation algorithm based on weighted trust relationship enhancement according to claim 1, characterized in that: The value range of λ is 0.275 to 0.

575.

7. The recommendation algorithm based on weighted trust relationship enhancement according to claim 6, characterized in that: In the FriendFeed dataset, λ is 0.375; in the Epinions dataset, λ is 0.275 or 0.

575.

8. The recommendation algorithm based on weighted trust relationship enhancement according to claim 1, characterized in that: If the target user u i Trust user u j , then b ij =1, otherwise b ij =0; when b ij = 0, the target user u i The final resource allocation obtained is expressed as: f ′(i) =S CosRA f (i) 。 9. The recommendation algorithm based on weighted trust relationship enhancement according to claim 1, characterized in that: The evaluation indicators of the accuracy of the algorithm include AUC, Ranking Score, nDCG, precision, recall rate and F1 index; the diversity indicators include one or more of Hamming distance, internal similarity and novelty.

10. The recommendation algorithm based on weighted trust relationship enhancement according to claim 1, characterized in that: The length of the recommendation list shall not exceed 50.

Images