Effective attack models for shilling item-based collaborative filtering systems

E?ective Attack Models for Shilling Item-Based

Collaborative Filtering Systems

Bamshad Mobasher,Robin Burke,Runa Bhaumik,Chad Williams

Center for Web Intelligence

School of Computer Science,Telecommunication and Information Systems

DePaul University,Chicago,Illinois

{mobasher,rburke,rbhaumik,cwilli}@https://www.wendangku.net/doc/1f1607510.html,

Abstract

Signi?cant vulnerabilities have recently been identi?ed in collaborative?ltering recommender systems.

These vulnerabilities mostly emanate from the open nature of such systems and their reliance on user-

speci?ed judgments for building pro?les.Attackers who cannot be readily distinguished from ordinary

users may introduce biased data in an attempt to force the system to“adapt”in a manner advantageous

to them.A handful of simple attack models have,so far,been identi?ed,and there appear to be

signi?cant di?erences in the susceptibility of di?erent recommendation techniques to these attacks.In

particular,item-based collaborative?ltering has been found to o?er some security advantages over user-

based collaborative?ltering.Our research in secure personalization is examining a range of more complex

attack models and recommendation techniques,paying particular attention to the costs and bene?ts of

mounting an attack.In this paper,we take a closer look at item-based collaborative?ltering.In

particular,we propose a new attack model that focuses on a subset of users with similar tastes and show

that such an attack can be highly successful against an item-based algorithm.

Key Words:Shilling,Collaborative Filtering,Recommender Systems,Attack Models

1.Introduction

Recent research has begun to examine the vulnerabilities and robustness of di?erent recommendation tech-niques,such as collaborative?ltering,in the face of what has been termed“shilling”attacks[2,1,5,6]. Attackers who cannot be readily distinguished from ordinary users may introduce biased data in an attempt to force the system to“adapt”in a manner advantageous to them.Recommendation systems,as well as many other user-adaptive systems are vulnerable to such attacks,precisely because they rely on users’in-teractions with the system and past user pro?les to generate recommendations or dynamic content.The wide-spread use of such systems in domains such as electronic commerce and information access provides a strong motivation for unscrupulous agents to use such attacks in the hope of gaining economic advantage.

It is easy to see why collaborative?ltering is vulnerable to shilling attacks.A user-based collaborative ?ltering algorithm collects user pro?les,which are assumed to represent the preferences of many di?erent individuals and makes recommendations by?nding peers with like pro?les.If the pro?le database contains biased data(many pro?les all of which rate a certain item highly,for example),these biased pro?les may be considered peers for genuine users and result in biased recommendations.This is precisely the e?ect found in[5]and[6].We have replicated these results and begun to extend them to consider alternative attack models.

Our work considers in particular the cost of mounting an attack.This cost has two primary components: knowledge cost and execution cost.Knowledge cost is the cost or e?ort required to gather information about the system to be attacked or its users.We assume that the more detailed the knowledge that is required by an attack(details of the rating distribution across pro?les,for example)the more costly the attack will be to mount.The execution cost is the e?ort required in terms of interactions with the system to add the necessary pro?les and ratings to execute the attack.While this latter cost may seem irrelevant

Figure1.The general form of a push attack pro?le.

when automated software agents can generate the needed pro?les,we believe that it remains a relevant consideration.To defend against shilling attacks,sites may implement policies limiting the speed with which pro?les can be built.Thus,an attack that requires a small number of short attack pro?les would be much more practical to mount and more di?cult to detect and defend against than one that requires that many pro?les be constructed,each with many ratings.

Lam et al.[5]show that item-based collaborative?ltering appears to o?er an advantage over the user-based approach.In item-based collaborative?ltering,the system looks for items with similar pro?les and makes predictions based on a user’s own rating of these peer items(see example below).By adding biased user pro?les,an attacker can only alter a portion of the pro?le for any given item.In[2],we suggested that an attack could be designed speci?cally to target an item-based recommendation algorithm if it is designed to change the distances between item pro?les in speci?c ways.This attack,called here the favorite item attack,is designed to target individual users by co-rating their favorite items with a target item.However, such an attack presents too signi?cant a knowledge requirement to be of practical use by an attacker.In order to know which items are the best peers for the target item,the attacker must know what each user’s ratings are for each item.

This paper proposes a generalization of the favorite item attack called the segmented attack.A segmented attack is one that pushes an item to a targeted group of users with known or easily predicted preferences. Pro?les are inserted that maximize the similarity between the pushed item and items preferred by the group.As we show below,the segmented attack is both e?ective and practical against standard item-based collaborative?ltering algorithms.

The paper is organized as follows.In Section2.we provide a detailed description of various attack models against collaborative?ltering systems,including those proposed in earlier work,as well as some that we have examined in out research.Section3.includes some background information and the speci?c details of the user-based and item-based recommendation algorithms used in our experiments.This section also contains a description of the evaluation metrics we have used to determine the e?ectiveness of various attack models. In Section4.we present our experimental results.We?rst show the impact of some of the previously studied attack models on the item-based algorithm.We then provide a detailed analysis of the proposed segmented attack model and experimentally show that it can be e?ective against item-based collaborative?ltering. 2.Attack Models

An attack against a collaborative?ltering recommender system consists of a set of attack pro?les,biased pro?le data associated with?ctitious user identities,and a target item,the item that the attacker wishes the system to recommend more highly(a push attack),or wishes to prevent the system from recommending(a nuke attack).We concentrate on push attacks in this paper.An attack model is an approach to constructing the attack pro?le,based on knowledge about the recommender system,its rating database,its products, and/or its users.The general form of a push attack pro?le is depicted in Figure1.An attack pro?le consists of an m-dimensional vector of ratings,were m is the total number of items in the system.The rating given to the pushed item,target,is r max and is the maximum allowable rating value.On the other hand,the ratings r1through r m?1are assigned to the corresponding items according to the speci?c attack model.Indeed,

Figure2.An example of a push attack favoring the target item Item6.

the speci?c strategy used to assign ratings to items1through m?1is what determines the type of attack model used.

In the remainder of this section,we provide an illustrative example that will help illustrate the vulner-ability of collaborative?ltering algorithms,and will serve as a motivation for the attack models,which we will then describe more formally.

2.1An Example

Consider,as an example,a recommender system that identi?es books that users might like to read using a user-based collaborative algorithm[3].A user pro?le in this hypothetical system might consist of that user’s ratings(in the scale of1-5with1being the lowest)on various books.Alice,having built up a pro?le from previous visits,returns to the system for new recommendations.Figure2shows Alice’s pro?le along with that of seven genuine users.An attacker,Eve,has inserted attack pro?les(Attack1-3)into the system,all of which give high ratings to her book labeled Item6.Eve’s attack pro?les may closely match the pro?les of one or more of the existing users(if Eve is able to obtain or predict such information),or they may be based on average or expected ratings of items across all users.

If the system is using a standard user-based collaborative?ltering approach,then the predicted ratings for Alice on Item6will be obtained by?nding the closest neighbors to Alice.Without the attack pro?les, the most similar user to Alice,using correlation-based similarity,would be User6.The prediction associated with Item6would be2,essentially stating that Item6is likely to be disliked by Alice.After the attack, however,the Attack1pro?le is the most similar one to Alice,and would yield a predicted rating of5for Item6,the opposite of what would have been predicted without the attack.So,in this example,the attack is successful,and Alice will get Item6as a recommendation,regardless of whether this is really the best suggestion for her.She may?nd the suggestion inappropriate,or worse,she may take the system’s advice, buy the book,and then be disappointed by the delivered product.

On the other hand,if a system is using an item-based collaborative?ltering approach,then the predicted rating for Item6will be determined by comparing the rating vector for Item6with those of the other items. This algorithm does not lend itself to an attack as obvious as the previous one,since Eve does not have control over ratings given by other users to any given item.However,if Eve can obtain some knowledge about the rating distributions for some items,this can make a successful attack more likely.In the example of Figure2,for instance,Eve knows that Item1is a popular item among a signi?cant group of users to which Alice also belongs.By designing the attack pro?les so that high ratings are associated with both Item1 and Item6,Eve can attempt to increase the similarity of these two items,resulting in a higher likelihood that Alice(and the rest of the targeted group)will receive Item6as a recommendation.Indeed,as the example portrays,such an attack is highly successful regardless of whether the system is using an item-based

Figure3.A Bandwagon attack pro?le.

or a user-based algorithm.This latter observation illustrates the motivation behind the attack model we introduce and analyze in this paper,namely the segmented attack.

2.2Attack Models

Prior work on recommender system stability has examined primarily three types of attack models:?Sampling attack:A sampling attack is one in which attack pro?les are constructed from entire user pro?les sampled from the actual pro?le database,augmented by a positive rating for the pushed item.

This attack is used by O’Mahony et al.[6]to provide a proof of the instability of collaborative?ltering algorithms,but is the least practical from a knowledge cost standpoint.

?Random attack:Lam et al.[5]show an attack model in which pro?les consist of random values (except of course for a positive rating given to the pushed item).Speci?cally,r1through r m?1are assigned to the corresponding items by generating random values within the rating scale with a dis-tribution centered around the mean for all user ratings across all items(see Figure1).The knowledge required to mount such an attack is quite minimal,especially since the overall rating mean in many systems can be determined by an outsider empirically(or,indeed,may be available directly from the system).The execution cost involved,however,is still substantial,since it involves assigning ratings to every item in each attack pro?le.Furthermore,as[5]shows and our results con?rm[1],the attack is not particularly e?ective.

?Average attack:A more powerful attack described in[5]uses the individual mean for each item rather than the global mean(except again the pushed item.)In the average attack,each assigned rating,r i,in an attack pro?le corresponds(either exactly or approximately)to the mean rating for item i,across the users in the database who have rated that item(see Figure1).In addition to the e?ort involved in producing the ratings,the average attack also has considerable knowledge cost of order m.Our experiments,however,have shown that,in the case of user-based algorithm,the average attack can be just as successful by assigning the average ratings to a small subset of items in the database,thus substantially reducing the knowledge requirement[1].This attack model,however,is not,in general,e?ective against an item-based collaborative algorithm,as show in Section4.below.

In addition to these attack models,we have introduced several others that are described below.Some of these attack models were introduced in[2]and were analyzed in the context of user-based collaborative ?ltering in[1].In this paper,we discuss these attacks in the context of item-based collaborative?ltering.

?Bandwagon attack:This attack takes advantage of the Zipf’s law distribution of popularity in consumer markets:a small number of items,best-seller books for example,will receive the lion’s share of attention and also ratings.The attacker using this model will build attack pro?les containing those items that have high visibility.Such pro?les will have a good probability of being similar to a large number of users,since the high visibility items are those that many users have rated.For example,by associating her book with current best-sellers,for example,The DaVinci Code,Eve can ensure that her bogus pro?les have a good probability of matching any given user,since so many users will have

Figure4.A Favorite Item attack pro?le.

these items on their pro?les.This attack can be considered to have low knowledge cost.It does not require any system-speci?c data,because it is usually not di?cult to independently determine what the“blockbuster”products are in any product space.

Figure3depicts a typical attack pro?le for the bandwagon attack.Items F R1through F R k are selected because they have been rated by a large number of users in the database.These items are assigned the maximum rating value together with the target item.The ratings r1through r m?k?1for the other items are determined randomly in a similar manner as in the random attack.The bandwagon attack therefore can be viewed as an extension of the random attack.We showed in[4]that the bandwagon attack can still be successful even when only a small subset of the“random items”,item1through item m?k?1are assigned ratings.However,as in the case of the average attack,it falls short when used against an item-based algorithm,as shown in Section4.below.

?Favorite item attack:(called the“consistency attack”in[2])Rather than knowledge about items, the favorite item attack looks at knowledge of user’s preferences.Such an attack is mounted not against the system as a whole,but by targeting a given user.We assume that the attacker knows which items a given user,u,really likes,and builds pro?les containing only those items.Like the sampling attack, this attack is not particularly practical from a knowledge cost standpoint,but provides an upper bound on the e?ectiveness of other attacks focused on user characteristics.

Figure4depicts a typical attack pro?le for the favorite item attack.F I i(u)represent the favorite items by user u selected in the attack pro?le.These favorite items are the ones whose ratings are greater than the user’s average rating.These items are assigned maximum rating value together with the target item.The other items in the database,item1through item m?k?1are assigned ratings at random or based on other criteria.In our experiments,best results were obtained when the non-favored items are assigned the lowest possible rating.Given its direct tailoring to each user,it is not surprising that the favorite item attack is e?ective against both user-based and item-based algorithms as our experiments have suggested[1].

?Segmented attack:The segmented attack is a generalization of the favorite item attack.It may be impossible to know what items are preferred by a given user,but it is possible to discover what items are well liked by a targeted segment of users and use this fact to attack that segment speci?cally.In fact,such an approach is probably one with great pragmatic appeal to an attacker.For example,if Eve were an author of a fantasy book for children,she would probably much prefer to have her book pushed to users who are fans of the Harry Potter series than to readers of gardening books.

This attack also requires very limited knowledge about the system and the users.An attacker needs to know only a group of items well liked by the target segment and needs to build pro?les containing only those items.Figure5depicts a typical attack pro?le for the segmented attack.Items SI1through SI k are the speci?c items,in our case they are the movies common to a segment of users.These items are assigned the maximum rating value together with the target item.The ratings r1through r m?k?1 are assigned ratings at random or based on other criteria.As with the favorite item attack,the best results were obtained when these items are assigned to1,the lowest possible rating.

Figure5.A Segmented attack pro?le.

3.Recommendation Algorithms and Evaluation Metrics

We have concentrated in this work on the most commonly-used algorithms for collaborative?ltering.Each algorithm assumes that there is a user/item pair for whom a prediction is sought,the target user and the target item.The task for the algorithm is to predict the target user’s rating for the target item.

3.1User-Based Collaborative Filtering

The standard collaborative?ltering algorithm is based on user-to-user similarity[3].This k NN algorithm operates by selecting the k most similar users to the target user,and formulates a prediction by combining the preferences of these users.k NN is widely used and reasonably accurate.The similarity between the target user,u,and a neighbor,v,can be calculated by the Pearson’s correlation coe?cient de?ned below:

sim u,v=

i∈I

(r u,i?ˉr u)?(r v,i?ˉr v)

i∈I

(r u,i?ˉr u)2?

i∈I

(r v,i?ˉr v)2

where I is the set of all items that can be rated,r u,i and r v,i are the ratings of some item i for the target user u and a neighbor v,respectively,andˉr u andˉr v are the average of the ratings of u and v over I, respectively.Once similarities are calculated,the most similar users are selected.In our implementation,we have used a value of20for the neighborhood size k.We also?lter out all neighbors with a similarity of less than0.1to prevent predictions being based on very distant or negative correlations.Once the most similar users are identi?ed,we use the following formula to compute the prediction for an item i for target user u.

p u,i=ˉr a+

v∈V

sim u,v(r v,i?ˉr v)

v∈V

|sim u,v|

where V is the set of k similar users and r v,i is the rating of those users who have rated item i,ˉr v is the average rating for the target user over all rated items,and sim u,v is the mean-adjusted Pearson correlation described above.The formula in essence computes the degree of preference of all the neighbors weighted by their similarity and then adds this to the target user’s average rating:the idea being that di?erent users may have di?erent“baselines”around which their ratings are distributed.

3.2Item-Based Collaborative Filtering

Item-based collaborative?ltering works by comparing items based on their pattern of ratings across users. Again,a nearest-neighbor approach can be used.The k NN algorithm attempts to?nd k similar items that are co-rated by di?erent users similarly.

For our purpose we have adopted the adjusted cosine similarity measure introduced by[7].The adjusted cosine similarity formula is given by:

sim i,j=

u∈U

(r u,i?ˉr u)?(r u,j?ˉr u)

u∈U

(r u,i?ˉr u)2?

n

u∈U

(r u,j?ˉr u)2

where r u,i represents the rating of user u on item i,andˉr u is the average of the user u’s ratings as before. After computing the similarity between items we select a set of k most similar items to the target item and generate a predicted value by using the following formula:

p u,i=

j∈J

r u,j?sim i,j

j∈J

sim i,j

where J is the set of k similar items,r u,j is the prediction for the user on item j,and sim i,j is the similarity between items i and j as de?ned above.We consider a neighborhood of size20and ignore items with negative similarity.The idea here is to use the user’s own ratings for the similar items to extrapolate the prediction for the target item.

3.3Evaluation Metrics

There has been considerable research in the area of recommender systems evaluation[4].Some of these concepts can also be applied to the evaluation of the security of recommender systems,but in evaluating security,we are interested not in raw performance,but rather in the change in performance induced by an attack.In[6]two evaluation measures were introduced:robustness and stability.Robustness measures the performance of the system before and after an attack to determine how the attack a?ects the system as a whole.Stability looks at the shift in system’s ratings for the attacked item induced by the attack pro?les.

Our goal is to measure the e?ectiveness of an attack-the“win”for the attacker.The desired outcome for the attacker in a“push”attack is of course that the pushed item be more likely to be recommended after the attack than before.In the experiments reported below,we follow the lead of[6]in measuring stability via prediction shift.However,we also measure hit ratio,the average likelihood that a top N recommender will recommend the pushed item[7].This allows us to measure the e?ectiveness of the attack on the pushed item compared to all other items.

Average prediction shift is de?ned as follows.Let U and I be the sets of target users and items,respec-tively.For each user-item pair(u,i)the prediction shift denoted by?u,i,can be measured as?u,i=p u,i?p u,i, where p represents the prediction after the attack and p before.A positive value means that the attack has succeeded in making the pushed item more positively rated.The average prediction shift for an item i over all users can be computed as

?i=

u∈U

?u,i/|U|.

Similarly the average prediction shift for all items tested can be computed as

ˉ?=

i∈I

?i/|I|.

Note that a strong prediction shift is not a guarantee that an item will be recommended-it is possible that other items’scores are a?ected by an attack as well or that the item scores so low to begin with that even a signi?cant shift does not promote it to“recommended”status.Thus,in order to measure the e?ectiveness of the attack on the pushed item compared to other items,we introduce the hit ratio metric.Let R u be the set of top N recommendations for user u.For each push attack on item i,the value of a recommendation

Figure6.Average Attack in Items-Based https://www.wendangku.net/doc/1f1607510.html,er-Based Collaborative Filtering.

hit for user u denoted by H ui,can be evaluated as1if i∈R u;otherwise H ui is evaluated to0.We de?ne hit ratio as the number of hits across all users in the test set divided by the number of users in the test set. The hit ratio for a pushed item i over all users in a set can then be computed as:

H ui/|U|.

HitRatio i=

u∈U

Likewise average hit ratio can then calculated as the sum of the hit ratio for each item i following an attack on i across all items divided by the number of items:

HitRatio i/|I|.

HitRatio=

i∈I

We plan to explore other metrics based on recommendation behavior,such as the bin-based techniques used in[5]and others,in our future work.

4.Experiments and Discussion

In our experiments we have used the publicly-available Movie-Lens100K dataset1.This dataset consists of 100,000ratings on1682movies by943users.All ratings are integer values between one and?ve where one is the lowest(disliked)and?ve is the highest(most liked).Our data includes all the users who have rated at least20movies.We used a neighborhood size of20in the k-nearest-neighbor algorithms for both item-based and user based techniques.To perform our attack experiments,we must average over a number of di?erent attack items,so we selected50movies taking care that the distribution of ratings for these movies matched the overall ratings distribution of all movies.We also generally selected a sample of50users as our test data, again mirroring the overall distribution of users in terms of number of movies seen and ratings provided. The results reported below represent averages over the combinations test users and test movies.We use the two metrics of prediction shift and hit ratio to measure the relative performance of various attack models. Generally,the values of these metrics are plotted against the size of the attack reported as the number of attack pro?les as a percentage of the total number of pro?les in the system.

Our earlier investigation[2,1],as well as the study reported in[5],suggest that while the average and random attacks can be successful against user-based collaborative systems,they generally fall short of having a signi?cant impact in the stability of item-based algorithms.For example,Figure6shows that item-based CF approach is more robust than the standard user-based algorithms in terms of the overall prediction shift on target items.Similar results were obtained when measuring the hit ratio.

1https://www.wendangku.net/doc/1f1607510.html,/research/GroupLens/data/

https://www.wendangku.net/doc/1f1607510.html,parison of Average and Bandwagon attacks in Item-Based algorithm.

In our earlier examination of user-based collaborative?ltering,we examined the bandwagon attack-a lower knowledge version of the random attack and found that it was comparable in performance to the average attack without requiring as much system-speci?c knowledge.We repeated these experiments against the item-based algorithm with similar results,namely that despite its lower knowledge requirements,the bandwagon attack was comparable to the average attack in impact.The average attack resulted in slightly higher in both prediction shift and hit ratio measures.However,the overall impact of this attack compared to the average attack was far less successful for the item-based algorithm con?rming the relative stability of the item-based algorithm over the user-based algorithm.These results are depicted in Figure7using10 Should we conclude then that an item-based algorithm is a successful defense against shilling attacks,or are there speci?c attack models that can have a practical impact on such systems?The favorite item attack was introduced in[2]as an approach that appeared to have a theoretical advantage over previously-developed attack types when applied to the item-based collaborative?ltering.Our preliminary results[1]showed that this attack model can be e?ective against both user-based and item-based algorithms.Here we extend those results and examine the segmented attack as a more e?ective and practical variation of the favorite item attack.

The favorite item attack assumes that we have knowledge of a handful of items that each user likes. Liked items are most likely to be rated-users can often predict that they will not like a particular movie and therefore avoid seeing it.Attack pro?les can then be assembled that consist of these liked items and the pushed movie.Other movies are assigned low ratings.Note that a new attack must be formulated for each target user.This is not practical,of course,but if we generalize from the single user to a market niche of users with similar tastes,it becomes plausible that an attacker might construct an attack targeted only to that niche.Indeed,the attacker might have demographic and marketing data that sorts the users into market segments whose preferences might be highly predictable.

Based on this observation we introduce the segmented attack(see Figure5),in which a set of items are selected for co-rating with the target item based on how they de?ne a segment of users.Thus,the segmented attack targets a set of users,in contrast to individual users in the favorite item attack.Furthermore,the selection of the segment is done implicitly(without direct knowledge about the individual users within the segment)by the virtue of selecting highly rated movies with similar characteristics.To build our segmented attack,we identi?ed a segment of users all of whom had given above average scores(4or5)to any three of the?ve movies,namely,Alien,Psycho,The Shining,Jaws,and The Birds.2

For this set of?ve movies,we then selected all combinations of three movies that had at least50users support,chose50of those users randomly and averaged the results.These results were also con?rmed with 2The list was generated from on-line sources of the popular horror?lms:https://www.wendangku.net/doc/1f1607510.html,/chart/horror and http://www.?https://www.wendangku.net/doc/1f1607510.html,/a?100thrillers1.html.

Figure8.Segmented attack in Item-Based algorithm

a di?erent segment based on Harrison Ford’s movies.The power of the segmented attack is emphasized in Figure8in which the impact of the attack is compared within the targeted user segment and within the set of all users.Left panel in the Figure shows the comparison in terms of prediction shift and varying attack sizes,while the right panel depicts the hit ratio at1%attack.While the segmented attack does show some impact against the system as a whole,it truly succeeds in its mission:to push the attacked movie precisely to those users de?ned by the segment.Indeed,in the case of in-segment users,the hit ratio is much higher than average attack.The chart also depicts the e?ect of hit ratio before any attack.Clearly the segmented attack has a bigger impact than any other attack we have previously examined against item-based algorithm. Our prediction shift results show that the segmented attack is more e?ective against in-segment users than even the more knowledge intensive average attack for the item-based collaborative algorithm.

Finally,we performed a set of experiments to determine the impact of“focus”on the segmented attack. By focus,here we mean the degree to which the user segment is characterized by its interest in a speci?c type of item.In this case we considered three user segments with increasing degrees of focus based on the movies they have rated highly.The?rst segment(focus1)consists of all those users who have given above average rating to any one of the?ve horror movies as mentioned above.The second segment(focus2)has users who have rated4or5any two of the?ve movies.Finally the users in the third segment(focus3)had rated above average any three of the?ve movies.As the focus increases,the user segments become smaller and increasingly characterized by those who enjoy horror movies.

We performed the segmented attack against each segment,in each case taking the movie combinations described above as the selected items that were co-rated with the target item in the attack pro?les.For the focus2and focus1results,the average was taken from running all combinations of the movies used in focus 3.The results of this experiment are depicted in Figure9,showing the prediction shift and hit ratio values, respectively,across the three segments.We?xed the number of top recommendations,N,to17(1%of the total items).As expected,the increase in focus results in the segmented attack having a higher impact in the targeted user segment,but even a segment de?ned by two liked movies in common is strongly impacted by the attack.

5.Conclusions

The open and interactive nature of collaborative?ltering is both a source of strength and vulnerability for recommender systems.As our research and that of others has shown,biased pro?le data can easily sway the recommendations of a collaborative system towards inaccurate results that serve the attacker’s ends.Previous research had held out hope that item-based collaborative?ltering might be relatively robust

Figure9.Analysis of in-segment focus in Segmented attack.

compared to the more common user-based variant.However,our research reported here shows that a fairly low-cost technique,the segmented attack,can be successfully deployed against item-based recommenders. Furthermore,the segmented attack o?ers pragmatic advantages for the attacker.Instead of spreading the bias due to the attacks across the whole user base,the segmented attack lets the attacker pick a focused set of users to whom an item should be pushed,e?ectively allowing targeted marketing of particular products to those sets of individuals judged as most likely to be in?uenced by the biased recommendation. References

[1]R.Burke,B.Mobasher,and R.Bhaumik.Limited knowledge shilling attacks in collaborative?lter-

ing systems.In Proceedings of the3rd IJCAI Workshop in Intelligent Techniques for Personalization, Edinburgh,Scotland,August2005.

[2]R.Burke,B.Mobasher,R.Zabicki,and R.Bhaumik.Identifying attack models for secure recommen-

dation.In Beyond Personalization:A Workshop on the Next Generation of Recommender Systems,San Diego,California,January2005.

[3]J.Herlocker,J.Konstan,A.Borchers,and J.Riedl.An algorithmic framework for performing collabora-

tive?ltering.In Proceedings of the22nd ACM Conference on Research and Development in Information Retrieval(SIGIR’99),Berkeley,CA,August1999.

[4]J.Herlocker,J.Konstan,L.G.Tervin,and J.Riedl.Evaluating collaborative?ltering recommender

systems.ACM Transactions on Information Systems,22(1):5–53,2004.

[5]https://www.wendangku.net/doc/1f1607510.html,m and J.Reidl.Shilling recommender systems for fun and pro?t.In Proceedings of the13th

International WWW Conference,New York,May2004.

[6]M.O’Mahony,N.Hurley,N.Kushmerick,and G.Silvestre.Collaborative recommendation:A robustness

analysis.ACM Transactions on Internet Technology,4(4):344–377,2004.

[7]B.Sarwar,G.Karypis,J.Konstan,and J.Riedl.Item-based collaborative?ltering recommendation

algorithms.In Proceedings of the10th International World Wide Web Conference,Hong Kong,May 2001.

三星通关秘籍!

三星通关秘籍！ 终于通关了!说说我的经验. 以下是我的阵容：

可见我的阵容并不强大,两个T都没满级,DPS也相当不给力.唯一特色是控比较多.这也是我能顺利通关的最重要的原因. 另外很多人也许会吐槽我的副T凤凰,之所以把它放在这个位置,是因为他的治疗技能,很多时候可以拯救世界.以下是他的技能: 凤凰冲击2级:攻击敌方全体目标,造成145%威力伤害,有几率触发集火 烈火精灵2级:召唤小凤凰，为自己以及队伍中血量最少的2名队友恢复25%血量 烈日炙烤0级:这个你要是学了,就废了. 下面请我的主力英雄隆重登场! 第一控:小小 山崩2级:攻击前排和中间群体敌人,造成120%威力伤害,并造成敌方眩晕,持续2秒 说明:此招一出,除了boss以外的所有野怪都被晕2秒! 每回合可释放次数:2次 第二控:流浪剑客 风暴之锤2级:攻击敌方中间竖排敌人,造成85%威力伤害,并造成敌方眩晕,持续2秒 说明:此招一出,boss和面前的2个野怪被晕2秒! 每回合可释放次数:8次 (15级的蓝色流浪剑客,每次耗蓝只有907) 第三控:萨尔 动能力场1级:攻击敌方所有目标,造成60%威力的伤害,并降低目标移动速度40%,持续5秒,有几率触发集火

说明:此招一出,所有怪减速40%,和流浪剑客的晕完美配合,很多时候怪都没有出手的机会! 每回合可释放次数:4次 既然控制这么给力,那么如何将控制的效率发挥到极致呢? 重要的有3点: 1. 要有速度! 以下是我的装备,可供大家参考. 值得一提的是那件”破损的远祖战鼓”.是可以和普通远祖战鼓完美叠加的,包括被动加速和主动加速,群体加速30有木有!

Tesla Model S底盘全透视..

水平对置、后置后驱、低重心、前双横臂后多连杆、全铝合金车架、5门5座，你以为笔者说的是保时捷新车型吗？那笔者再补充多几个关键词好了，后置的水平对置双电刷电动机、0油耗、藏在地板下的笔记本电池组，同时拥有这些标签的，便是Tesla第二款车型Model S。Model S是五门五座纯电动豪华轿车，布局设计及车身体积与保时捷Panamera相当，并且是目前电动车续航里程的纪录保持者（480公里）。虽然现在纯电动在我国远未至于普及，但是在香港地区却是已经有Tesla的展厅，在该展厅内更是摆放了一台没有车身和内饰，只有整个底盘部分的Model S供人直观了解Model S的技术核心。 图：Tesla Model S。

图：拆除车壳之后，Model S的骨架一目了然。

图：这套是Model S的个性化定制系统，可以让买家选择自己喜爱的车身颜色、内饰配色和轮圈款式，然后预览一下效果。可以看到Model S共分为普通版、Sign at ure版和Performance版，后面两个型号标配的是中间的21寸轮圈，而普通版则是两边的19寸款式。Signature版是限量型号，在美国已全部售罄，香港也只有少量配额。 图：笔者也尝试一下拼出自己心目中的Model S，碳纤维饰条当然是最爱啦。

图：参观了一下工作车间，不少Roadster在等着检查保养呢，据代理介绍，不同于传统的汽车，电动车的保养项目要少很多，至少不用更换机油和火花塞嘛，换言之电动车的维护成本要比燃油汽车要低。 Tesla于2010年5月进军香港市场，并于翌年2011年9月成立服务中心。由于香港政府对新能源车的高度支持，香港的电动车市场发展比起大陆地区要好得多。例如Tesla的第一款车型Roadster（详见《无声的革命者——Tesla Roadster Sport 》），在香港获得豁免资格，让车主可以节省将近100万港元的税款。在这样的优惠政策之下，Tesla Roadster尽管净车价达100万港元，但50台的配额已经基本售罄。而Model S目前在香港已经开始接受报名预定，确定车型颜色和配置之后约两个月左右可以交车。

特斯拉电动汽车动力电池管理系统解析(苍松书屋)

特斯拉电动汽车动力电池管理系统解析 1. Tesla目前推出了两款电动汽车，Roadster和Model S，目前我收集到的Roadster 的资料较多，因此本回答重点分析的是Roadster的电池管理系统。 2. 电池管理系统（Battery Management System, BMS）的主要任务是保证电池组工作在安全区间内，提供车辆控制所需的必需信息，在出现异常时及时响应处理，并根据环境温度、电池状态及车辆需求等决定电池的充放电功率等。BMS的主要功能有电池参数监测、电池状态估计、在线故障诊断、充电控制、自动均衡、热管理等。我的主要研究方向是电池的热管理系统，因此本回答分析的是电池热管理系统 (Battery Thermal Management System, BTMS). 1. 热管理系统的重要性 电池的热相关问题是决定其使用性能、安全性、寿命及使用成本的关键因素。首先，锂离子电池的温度水平直接影响其使用中的能量与功率性能。温度较低时，电池的可用容量将迅速发生衰减，在过低温度下（如低于0°C）对电池进行充电，则可能引发瞬间的电压过充现象，造成内部析锂并进而引发短路。其次，锂离子电池的热相关问题直接影响电池的安全性。生产制造环节的缺陷或使用过程中的不当操作等可能造成电池局部过热，并进而引起连锁放热反应，最终造成冒烟、起火甚至爆炸等严重的热失控事件，威胁到车辆驾乘人员的生命安全。另外，锂离子电池的工作或存放温度影响其使用寿命。电池的适宜温度约在10~30°C之间，过高或过低的温度都将引起电池寿命的较快衰减。动力电池的大型化使得其表面积与体积之比相对减小，电池内部热量不易散出，更可能出现内部温度不均、局部温升过高等问题，从而进一步加速电池衰减，缩短电池寿命，增加用户的总拥有成本。 电池热管理系统是应对电池的热相关问题，保证动力电池使用性能、安全性和寿命的关键技术之一。热管理系统的主要功能包括：1）在电池温度较高时进行有效散热，防止产生热失控事故；2）在电池温度较低时进行预热，提升电池温度，确保低温下的充电、放电性能和安全性；3）减小电池组内的温度差异，抑制局部热区的形成，防止高温位置处电池过快衰减，降低电池组整体寿命。 2. Tesla Roadster的电池热管理系统 Tesla Motors公司的Roadster纯电动汽车采用了液冷式电池热管理系统。车载电池组由6831节18650型锂离子电池组成，其中每69节并联为一组（brick），再将9组串联为一层（sheet），最后串联堆叠11层构成。电池热管理系统的冷却液为50%水与50%乙二醇混合物。

三星S6E手机完全精简攻略

Galaxy S6/S6E完全精简攻略 1、【绝对不能删】删除后系统无法正常运行的应用 /system/app下： BadgeProvider（应用程序脚标服务，删除后不停报错） InCallUI（通话过程服务，删除了无法使用任何通话**能，也无法挂断电话） SimCardMgr（双卡管理服务，删除了无法正常通讯） /system/priv-app下： DefaultContainerService（应用容器，系统基础服务） ExternalStorageProvider（存储器，系统基础服务） LogsProvider（通话记录服务，删了不能打电话） MtpApplication（USB连接服务，删了连不了电脑，还会报错） SecMediaProvider（存储器，系统基础服务） SecSettings2.apk（设置，系统基础服务） SecSettingsProvider2.apk（设置，系统基础服务） InputDevices（输入服务，删了自己想吧） SecContacts_L_Phone_FLAGSHIP_CHN（联系人，删了无法管理联系人） SecContactsProvider（联系人存储，删了无法存储联系人） SharedStorageBackup（共享存储备份，系统基础服务） SystemUI（系统UI，系统基础服务） Telecom（电话，删了没信号） TeleService（通讯基础服务，删了没信号） 简单点说，以上东西就是S6上最小的系统，包括数据连接在内的核心通讯功能全部正常。缺点就是众多**功能缺失，但如果你就打电话发短信，只用几个固定应用，不拍照不用蓝牙，且事先已经安装好常用APP（如微信、浏览器什么的），那么也就够用了，适合疯狂追求最简的人们。这里需要提醒的是，我列出的只是文件夹形式存在的App，在/system/app目录下还有一个文件名很长的单独文件（忘了名字，明天查查再补充），那个千万不能删，删了直接就无法引导了。 2、【强烈建议不要删】可以删、但删除会导致较为严重后果的应用： /system/app下： Bluetooth（顾名思义，删了就用不了蓝牙） NfcNci（NFC服务，删了NFC就挂了，不过这个用的人应该很少） mcRegistry（删除后wifi开关巨慢、不能存储密码且系统性能会下降） PackageInstaller（应用安装服务，删了安装不了应用） PacProcessor（应用处理服务，删了安装不了应用） SecHTMLViewer（HTML浏览服务，很多应用需要调用这个才能正常现实HTML内容） WebViewGoogle（新版HTML浏览服务，和上面那个功能一致，但使用的是Webkit核心，很多新应用调用的是这个）

特斯拉整体介绍

Tesla Model S 特斯拉Model S是一款纯电动车型，外观造型方面，该车定位一款四门Coupe车型，动感的车身线条使人过目不忘。此外在前脸造型方面，该车也采用了自己的设计语言。另值得一提的是，特斯拉Model S的镀铬门把手在触摸之后可以自动弹出，充满科技感的设计从拉开车门时便开始体现。该车在2011年年中正式进入量产阶段，预计在2012年年内将有5000台量产车投放市场。 目录 1概述 2售价 3内饰 4动力 5车型 6技术规格 7性能表现 8荣誉 9对比测试 10车型参数 1概述

Tesla Model S是一款由Tesla汽车公司制造的全尺寸高性能电动轿车，预计于2012年年中投入销售，而它的竞争对手则直指宝马5系。该款车的设计者Franz von Holzhausen，曾在马自达北美分公司担任设计师。在Tesla汽车公司中，Model S拥有独一无二的底盘、车身、发动机以及能量储备系统。Model S的第一次亮相是在2009年四月的一期《大卫深夜秀》节目中 4 Tesla Model S 。 2售价 Model S的电池规格分为三种，分别可以驱动车辆行驶260公里、370公里和480公里。而配备这三种电池的Model S的售价则分别为57400美元、67400美元和77400美元。下线的首批1000辆签名款车型将配有可以行驶480公里的蓄电池。尽管官方尚未公布该签名款车型的具体售价，但据推测，价格将会保持在50000美元左右。 Tesla汽车公司称其将会对市场出租可以提供480公里行驶距离的电池。而从Model S中取得的收益将为第三代汽车的发展提供资金保障。 3内饰

基于4P-4C-4R理论的特斯拉电动汽车品牌营销策略探究

基于4P-4C-4R理论的特斯拉电动汽车品牌营销策略探究 兰州大学管理学院：鲍瑞雪指导老师：苏云【摘要】：特斯拉汽车作为新兴的电动汽车品牌，其发展速度是十分惊人的，尤其是自2014年进入中国市场后更是在中国新能源汽车业引起巨大的轰动，其成功模式值得深思，其中特斯拉汽车公司的营销策略，不仅根据目标市场进行分析，还结合了自身的实际情况，通过对多种营销策略的综合利用，最终实现了成功，促进了企业的持续发展。本文基于营销学4P-4C-4R理论对特斯拉汽车公司的营销策略进行深入分析，研究表明特斯拉从顾客角度出发，采取4P、4C、4R相结合的营销策略。4P是战术，4C是战略，应用4C、4R来思考， 4P角度来行动。 【关键词】： 4P 4C 4R 品牌营销策略特斯拉电动汽车 一、研究背景 当前，美国电动汽车企业特斯拉公司已经成为世界电动汽车界的一匹黑马。特斯拉公司成立于2003年，不同于其它新能源汽车企业，它主要从事电动跑车等高端电动汽车的设计、制造和销售。成立仅仅10年，特斯拉就已经开始在新能源汽车企业和高端汽车企业中崭露头角。作为新兴电动汽车品牌，特斯拉汽车的成功不是偶然的，这与其注重品牌营销策略密不可分。特斯拉汽车公司不仅对目标市场进行具体细分，结合自身实际进行有特色的精准的定位，还充分利用名人效应，提升品牌形象，并借助已有成功模式，寻求找到符合自身发展的途径，出一条饱含自己特色的品牌营销之路。 二、理论依据及文献综述 （一）理论依据 本文以经典的营销学理论4P-4C-4R理论为理论依据，着力探究4P-4C-4R 理论与特斯拉电动汽车营销策略之间的内在联系，分析特斯拉电动汽车品牌营销策略的经典之处。 1、4P营销理论 4P营销理论被归结为四个基本策略的组合，即产品（Product）、价格（Price）、渠道（Place）、促销（Promotion）。该理论注重产品开发的功能，要求产品有独特的卖点，把产品的功能诉求放在第一位。在价格策略方面，企业应当根据不

愤怒的小鸟：里约大冒险全三星图文攻略

愤怒的?鸟：?约?冒险--全三星图?攻略 三星要点： 1.尽量?最少的?鸟来完成任务 2.攻击?标应该为障碍物的薄弱点，譬如?柱 3.利??些附加的障碍物来达到摧毁其他建筑物的?的 S M U G G L E R'S D E N 1-1 第?关很简单，你所使?的是三个红??鸟;只需调整?向击中右部的中间?柱就可以了 1-3 如下图，射击的?标瞄准障碍物的下部。同时要让你的红??鸟在最后击中最右边的笼?，然后让整个障碍物失去平衡往左边下落。??个?鸟完成任务即可三星

1-5 这?关也是，让红??鸟摧毁左边的笼?;然后让其下落的时候击中右边的障碍物下部来让右边的建筑物失去平衡。 1-6 这?关会出现新的?鸟，蓝??鸟。这种?鸟的威?在于，它们?次可以射出三只?鸟，你可以让它们各个突破，击中?系列?标，但最理想的状态是，让它们三者近距离出击同?个?标，发挥最?攻击?。 射出之后再点击?次?标左键即可分为三只?鸟

这?关很简单，按照如下的路线在半程散射即可完成本任务 1-7 这?关要在快接近?标的时候散射，同时要确保三个?鸟有?少?个击中中间障碍物的最下部;让其失去平衡从?完成任务

1-8 这?关有相当的技巧，路线如下。两个?鸟攻击下?的障碍物，然后?个?鸟将最右上的?头击落从?让其下落把所有的笼?砸开 1-9 这?关还是散射，注意要保证三个?鸟能够击中障碍物的下部

1-10 这?关的?的就是让你彻底摧毁前?两个建筑物，然后让其剩余的物件如同多?诺?牌?样把后?的建筑物推到完成任务 1-11 本关会开启新的黄??鸟。这种?鸟虽然极其神速，但碰到?壁和玻璃墙时的攻击?仍然有限。?这种?鸟进攻?头壁垒简直像捅破?层纸?样好?，这样你就可以闪电袭击敌?，让它们葬?于堡垒的废墟之中。?且在发射之后你还可以按?次?标左键进?加速攻击

详解特斯拉Model S

详解特斯拉Model S 1、Model S的核心技术是什么？ 核心技术是软件，主要包括电池管理软件，电机以及车载设备的电控技术。最重要的是电池控制技术。 Model S的加速性能，续航里程、操控性能的基础都是电池控制技术，没有电池控制技术，一切都就没有了。 2、Model S的电池控制技术有什么特色？ 顶配的Model S使用了接近7000块松下NCR 18650 3100mah电池，对电池两次分组，做串并联。设置传感器，感知每块电池的工作状态和温度情况，由电池控制系统进行控制。防止出现过热短路温度差异等危险情况。 在日常使用中，保证电池在大电流冲放电下的安全性。 其他厂商都采用大电池，最多只有几百块，也没有精确到每块电池的控制系统。 3、为什么要搞这么复杂的电池控制系统？ 为了能够使用高性能的18650钴锂电池。高性能电池带来高性能车。因为18650钴锂电池的高危性，没有一套靠谱的系统，安全性就不能保证。这也是大多数厂商无论电力车，插电车，混合动力车都不太敢用钴锂电池，特别是大容量钴锂电池的原因。 松下NCR 18650 3100mah，除了测试一致性最好，充放电次数多，安全性相对较好以外，最重要的是能量大，重量轻，价格也不高。 由于能量大，重量轻，在轿车2吨以内的车重限制下，可以塞进去更多的电池，从而保证更长的续航里程。因为电池输出电流有限制，电池越多，输出电流越大，功率越大，可以使用的电机功率也就越大。电机功率越大，相当于发动机功率大，车就有更快的加速性能，而且可以保持较长的一段时间。 4、作为一辆车，Model S有哪些优点？这些优点是电动车带来的吗？ 作为一辆车，Model S主要具有以下几个优点 （1）起步加速快，顶配版本0-100公里加速4秒多，能战宝马M5

魔灵召唤三星魔灵最佳选择及培养攻略

魔灵召唤三星魔灵最佳选择及培养攻略 魔灵召唤三星魔灵魔灵召唤三星魔灵怎么选择培养，想了解一下更多有关魔灵召唤三星魔灵魔灵召唤三星魔灵怎么选择培养最新的攻略及资讯吗？下面就由小骨来为大家带来魔灵召唤三星魔灵魔灵召唤三星魔灵怎么选择培养的推荐 魔灵召唤三星魔灵最佳选择及培养建议。今天一起来看下大神给小白们的建议吧，尤其是前期可选择的三星魔灵的相关讲义，和一些中后期有潜力的三星魔灵推荐，魔灵召唤这款游戏非常的有内涵，需要小伙伴们深入的去体验才能更好的游戏，属于慢热经久耐玩的游戏类型。 火系： 1，火犬神 火犬为什么是新手第一选择狗粮队长，除了它的技能优势，被动是消灭敌人后获得额外攻击次数。也是因为相对容易升技能，得到渠道也简单，火山可以刷到，水犬，风犬前面地图就有，很容易满技能，付文不好可以带沙漠，付文好就可以带火山，当然你有好的胎4.胎5就另说了，不然就是平民第一狗粮队长选择。数据请自己查找，一般五星犬神，付文一般的带沙漠3，付文好带普通火山，六星好付文可带地狱沙漠，或者困难火山!注意普通火山速度需要110以上。另外犬神，竞技场，爬塔等其他地方表现也很好!付文，猛攻刀刃，2.46攻击，或者2.6攻击，4爆率，堆点速度! 2，火死神 这个也是狗粮队长，但它是中后期，因为前期比较难得到，全靠抽取，并且技能难升，而且付文不好时候并不一定比犬神快!但技能，付文，上6星，带困难，地狱火山都是一把好手!付文参考犬神!选择升5，有兴趣才6! 3，火地狱火 这个在遗迹有掉，也可以作为狗粮队长，但它的优势在于2技能满后，群体90%破甲，是前期三星唯一一个高概率群破甲!中期没水海盗，水偷，也可用他代替，3技能群晕一回合也不错!如果

TESLA特斯拉解析

TESLA 硅谷工程师、资深车迷、创业家马丁·艾伯哈德（Martin Eberhard）在寻找创业项目时发现，美国很多停放丰田混合动力汽车普锐斯的私家车道上经常还会出现些超级跑车的身影。他认为，这些人不是为了省油才买普锐斯，普锐斯只是这群人表达对环境问题的方式。于是，他有了将跑车和新能源结合的想法，而客户群就是这群有环保意识的高收入人士和社会名流。 2003年7月1日，马丁·艾伯哈德与长期商业伙伴马克·塔彭宁（Marc Tarpenning）合伙成立特斯拉（TESLA）汽车公司，并将总部设在美国加州的硅谷地区。成立后，特斯拉开始寻找高效电动跑车所需投资和材料。

由于马丁·艾伯哈德毫无这方面的制造经验，最终找到AC Propulsion公司。当时，对AC Propulsion公司电动汽车技术产生兴趣的还有艾龙·穆思科（Elon Musk）。在AC Propulsion公司CEO汤姆·盖奇（Tom Gage）的引见下，穆思科认识了艾伯哈德的团队。2004年2月会面之后，穆思科向TESLA投资630万美元，但条件是出任公司董事长、拥有所有事务的最终决定权，而艾伯哈德作为创始人任TESLA的CEO。 在有了技术方案、启动资金后，TESLA开始开发高端电动汽车，他们选择英国莲花汽车的Elise作为开发的基础。没有别的原因，只是因为莲花是唯一一家把TESLA放在眼里的跑车生产商。

艾伯哈德和穆思科的共同点是对技术的热情。但是，作为投资人，穆思科拥有绝对的话语权，随着项目的不断推进，TESLA开始尝到“重技术研发轻生产规划、重性能提升轻成本控制”的苦果。2007年6月，离预定投产日期8月27日仅剩下两个月时，TESLA还没有向零部件供应商提供Roadster的技术规格，核心的部件变速箱更是没能研制出来。另一方面，TESLA在两个月前的融资中向投资人宣称制造Roadster的成本为6.5万美元，而此时成本分析报告明确指出Roadster最初50辆的平均成本将超过10万美元。 生意就是生意，尤其硅谷这样的世界级IT产业中心，每天都在发生一些令人意想不到的事情。投资人穆思科以公司创始人艾伯哈德产品开发进度拖延、成本超支为由撤销其

全系别增益整理与简析

全系别增益整理与简析 恶魔系： 实用性：★★★★★ 作用：全员加180护甲。赤裸裸的土豪系，180护甲给谁都好用！ 野兽系： 实用性：★★★☆☆ 作用：全员加182力量。看似实用的属性，给了四个奇葩的前排壮汉… 巨人系： 实用性：★★★★★ 作用：全员加180魔抗。同样是前排巨汉，比野兽系的几个强悍太多。 斧系： 实用性：★★☆☆☆ 作用：全员加180力量 这是一个极其混乱的系别，宙斯也算斧系… 巨魔系： 实用性：★★★☆☆ 作用：提供生命回复。巨魔这个曾经的竞技场主角，能不能翻身？我感觉有点悬。 飞行系： 实用性：★★★★☆ 作用：全员加810法强。这是一个版面输出很高的团队技能，唯一的缺点是很难组合。

鱼人系： 实用性：★★★★☆ 作用：全员加180敏捷。非常好的一个技能，只可惜大鱼人连个闪避都没有，还站位那么靠前… 近卫系： 实用性：★★★★☆ 作用：全员加3960生命值。五小强中的三个着实火了一阵，相信奇迹相信梦想！ 刀剑系： 实用性：★★★☆☆ 作用：全员加180点物暴。唯一能输出的仅有剑圣和火猫，且都难堪主T 位置，遗憾！ 机械系： 实用性：★★★★☆ 作用：全员加27％的攻速和施法。有人说飞机是逗比，没错，但这确实是个好技能啊。（等待伐木机和发条登场咯） 西行系： 实用性：★★★☆☆☆ 作用：全员加72点魔抗。猴子强，则西行强，反之亦然。猴子和刚被的觉醒决定了这支队伍的未来走向。 弓箭系： 实用性：★★★★★☆ 作用：全员加36.4命中。受益最大的自然是一姐，信一姐得永生！ 总结： 英雄的系别构成是游戏非常重要的一个环节，因为仅仅需要两个同系别的英雄就能形成不错的收益，相当多了一件非常不错的高级装备。而这些属性的获得方式也比洗练来得更为简单和划算

水太深了,防骗必看!史上最全的三星手机盖世S3 i9300鉴定攻略-购买指南(超详细)

水太深了,防骗必看！史上最全的三星S3 i9300鉴定攻略-购买指南（超详细） 这是我个人买9300手机时参考用的，推荐 大家下载 首先公布两个容易造成误导的网站，这两个网站可以查询手机的型号和主板，查询销售地无效，在这两个网站上面查不到也不代表是假的： https://www.wendangku.net/doc/1f1607510.html,/ https://www.wendangku.net/doc/1f1607510.html,/?page=analysis&sub=imeinr 另外，看串号第几位来识别版本的方法就更离谱了，10年前就不可靠了。 言归正传： 先看主机： 一：看外观，成色怎么看就不用细说了，你懂的。以下主要讲三点:1：标签：看S/N码。S/N 码第4、5位分别代表出厂年份和月份。8代表08年，9代表09年，A代表10年，B代表11年，C代表2012年。第5位代表月份。如图这台手机，C7就说明是2012年7月份出厂的（电池上的S/N 码同理）。另外说明：港版的标签是中文的， ps:最新产的港版白色，电池仓标签，已经不是made by sumsang，而是直接“三星电子（株）韩国制造”，如图

2：看螺丝标：港版有蓝色螺丝标，写着原装标贴，撕毁无效。如图

3：看螺丝，估计有人问我发个螺丝干嘛？？其实我是回答你们最关心的问题“如何鉴别翻新机”，以三星为例，你们把手上的机子，随便拆一个螺丝出来看，如果螺丝上面有上图那样有油漆式的痕迹，那么恭喜你，是全新机，如果是全光滑的表面，那么是被换过的螺丝，翻新几率99%，这时候你们会质疑了，这么一个螺丝可以断定，我可以告诉你，这是行业内最直接最简单的鉴别方式，市场回收机都是这样检验的，信不信，你就拆一个螺丝看看，如果是光滑的，那么建议你去拆机验证，如果不是我说的，我可以帮你出拆机费，好了，我又曝光了一个行业内幕，大家可以试试，就一个螺丝，对手机没影响。如图

特斯拉电动汽车电池管理系统解析

1. Tesla目前推出了两款电动汽车，Roadster和Model S，目前我收集到的Roadster的资料较多，因此本回答重点分析的是Roadster的电池管理系统。 2. 电池管理系统（Battery Management System, BMS）的主要任务是保证电池组工作在安全区间内，提供车辆控制所需的必需信息，在出现异常时及时响应处理，并根据环境温度、电池状态及车辆需求等决定电池的充放电功率等。BMS的主要功能有电池参数监测、电池状态估计、在线故障诊断、充电控制、自动均衡、热管理等。我的主要研究方向是电池的热管理系统，因此本回答分析的是电池热管理系统 (Battery Thermal Management System, BTMS). 1. 热管理系统的重要性 电池的热相关问题是决定其使用性能、安全性、寿命及使用成本的关键因素。首先，锂离子电池的温度水平直接影响其使用中的能量与功率性能。温度较低时，电池的可用容量将迅速发生衰减，在过低温度下（如低于0°C）对电池进行充电，则可能引发瞬间的电压过充现象，造成内部析锂并进而引发短路。其次，锂离子电池的热相关问题直接影响电池的安全性。生产制造环节的缺陷或使用过程中的不当操作等可能造成电池局部过热，并进而引起连锁放热反应，最终造成冒烟、起火甚至爆炸等严重的热失控事件，威胁到车辆驾乘人员的生命安全。另外，锂离子电池的工作或存放温度影响其使用寿命。电池的适宜温度约在10~30°C 之间，过高或过低的温度都将引起电池寿命的较快衰减。动力电池的大型化使得其表面积与体积之比相对减小，电池内部热量不易散出，更可能出现内部温度不均、局部温升过高等问题，从而进一步加速电池衰减，缩短电池寿命，增加用户的总拥有成本。 电池热管理系统是应对电池的热相关问题，保证动力电池使用性能、安全性和寿命的关键技术之一。热管理系统的主要功能包括：1）在电池温度较高时进行有效散热，防止产生热失控事故；2）在电池温度较低时进行预热，提升电池温度，确保低温下的充电、放电性能和安全性；3）减小电池组内的温度差异，抑制局部热区的形成，防止高温位置处电池过快衰减，降低电池组整体寿命。 2. Tesla Roadster的电池热管理系统 Tesla Motors公司的Roadster纯电动汽车采用了液冷式电池热管理系统。车载电池组由6831节18650型锂离子电池组成，其中每69节并联为一组（brick），再将9组串联为一层（sheet），最后串联堆叠11层构成。电池热管理系统的冷却液为50%水与50%乙二醇混合物。 图 1.(a)是一层（sheet）内部的热管理系统。冷却管道曲折布置在电池间，冷却液在管道内部流动，带走电池产生的热量。图 1.(b)是冷却管道的结构示意图。冷却管道内部被分成四个孔道，如图 1.(c)所示。为了防止冷却液流动过程中温度逐渐升高，使末端散热能力不佳，热管理系统采用了双向流动的流场设计，冷却管道的两个端部既是进液口，也是出液口，如图 1(d)所示。电池之间及电池和管道间填充电绝缘但导热性能良好的材料（如Stycast 2850/ct），作用是：1)将电池与散热管道间的接触形式从线接触转变为面接触；2)有利于提高单体电池间的温度均一度；3)有利于提高电池包的整体热容，从而降低整体平均温度。

《使命召唤6现代战争2》特别任务单人3星全关卡心得攻略

《使命召唤6现代战争2》特别任务单人3星全关卡心得攻略 前言： 本人是个FPS苦手，基本不太玩FPS，大部分FPS都会晕，COD4，COD5也会晕， 但COD6的特别任务太好玩了，欲罢不能（COD6居然完全不晕了，哈哈） 写了点心得，希望帮助跟自己一样苦手或更苦手的玩家，别半途而废 特别任务单人3星看似很难但都有方法过去的哈哈，我都过了，你也可以的 只要用心，一定能过的，诀窍一个字-----“心”....呵呵 ----------------------------------------------------- 提示： 先玩1星难度熟悉流程寻找套路，再攻3星会好打很多 每关开始研究地图，研究地形，研究敌人的规律 多利用屏幕左上的小地图，敌人开枪时会显示位置 有时可以跳一跳，打几枪引敌 地雷是好帮手，插在敌人必经路看不见的拐角有大用 只有一条命，所以慢慢摸索，以静制动，敌人会经常来找你或绕后偷袭你的 ----------------------------------------------------- 注意： 为了不影响你自己的过关乐趣，最好先不要看，一旦看了将严重影响你通关的乐趣 游戏的乐趣在于自己攻克难关，达成3星时的兴奋与激动，所以慎重观看 实在苦手，实在过不去再看，不然你会后悔的，知道方法后将缺失了很多乐趣 ----------------------------------------------------- ■1-1-★★★训练关 多练练，边跑边打，用手枪m9，或mp5好点，熟练是关键 ----------------------------------------------------- ■1-2-★★★高台防守耍耍狙 第1波敌人杀到留下最后1个，然后下去四周拐角处埋点地雷，高台上也埋点 每波开局后的控制飞弹是关键，尽量炸多的人和一开始来的车，车上会下人 时刻注意是否有敌人靠近（用显示敌人位置的枪） ----------------------------------------------------- ■1-3-★★★巴西小巷好多人 开局拿到左前方凳子上有带显示敌人位置的枪是关键，稳扎稳打，注意会有敌人包抄绕后 某些地方埋点地雷防绕后敌人和狗 ----------------------------------------------------- ■1-4-★★★雪地谜踪 开局走右上方躲过第1波敌人，其后简单，有些地方可以爬在地上过去不会被发现，不难

Tesla Model S电池组设计全面解析

Tesla Model S电池组设计全面解析 对Tesla来说最近可谓是祸不单行；连续发生了3起起火事故，市值狂跌40亿，刚刚又有3名工人受伤送医。Elon Musk就一直忙着到处“灭火”，时而还跟公开表不对Tesla“不感冒”的乔治·克鲁尼隔空喊话。在经历了首次盈利、电池更换技术·穿越美国、水陆两栖车等头条新闻后，Elon Musk最近总以各种负面消息重返头条。这位"钢铁侠。CE0在201 3年真是遭遇各种大起大落。 其中最为人关注的莫过于Model S的起火事故，而在起火事故中最核心的问题就是电池技术。可以说，牵动Tesla股价起起落落的核心元素就是其电池技术，这也是投资者最关心的问题。在美国发生的两起火事故有着相似的情节Model S 撞击到金属物体后，导致电池起火，但火势都被很好地控制在车头部分。在墨西哥的事故中，主要的燃烧体也是电池；而且在3起事故中，如何把着火的电池扑灭对消防员来说都是个难题。 这让很多人产生一个疑问：Model S的电池就这么不禁撞吗?在之前的一篇文章中，我跟大家简单讨论了一下这个问题，但只是停留在表面。读者普遍了解的是，Model S的电池位于车辆底部，采用的是松下提供的18650钴酸锂电池，整个电池组包含约8000块电池单元；钴酸锂电池能量密度大，但稳定性较差，为此Tesla研发了3级电源管理体系来确保电池组正常运作。现在，我们找到了Tesla的一份电池技术专利，借此来透彻地了解下Model S电池的结构设计和技术特征。 电池的布局与形体

FIG3 如专利图所示，Model S的电池组位于车辆的底盘，与轮距同宽，长度略短于轴距。电池组的实际物理尺寸是：长2.7m，宽1.5m，厚度为0.1 m至0.1 8m。其中0.1 8m较厚的部分是由于2个电池模块叠加而成。这个物理尺寸指的是电池组整体的大小，包括上下、左右、前后的包裹面板。这个电池组的结构是一个通用设计，除了18650电池外，其他符合条件的电池也可以安装。此外，电池组采用密封设计，与空气隔绝，大部分用料为铝或铝合金。可以说，电池不仅是一个能源中心，同时也是Model S底盘的一部分，其坚固的外壳能对车辆起到很好的支撑作用。 由于与轮距等宽，电池组的两侧分别与车辆两侧的车门槛板对接，用螺丝固定。电池组的横断面低于车门槛板。从正面看，相当于车门槛板"挂着。电池组。其连接部分如下图所示。 FIG, 4

愤怒的小鸟星战版——图文全三星

愤怒的?鸟:星战版——图?全三星 昨天，R o v i o再次为我们奉献上了《愤怒的?鸟》最新版《星球?战》，操作?法还是?样，但各种鸟类有了新的功能，披上了星际?战的华丽外?，红鸟会挥舞光剑、黄鸟能发射激光等，当然，猪猪们也都进化了，不再是单独的被动挨打了，炮塔能主动向?鸟发动攻击了，?鸟们将?临着?次巨?的挑战，下?是?编为?家准备的全三星图?攻略。 如果你想在电脑上玩的话，请下载《愤怒的?鸟星球?战》P C版，已经破解和汉化。 《愤怒的?鸟星球?战》汉化破解版： h t t p://w w w.962.n e t/y o u x i/12414.h t m l 1-1： 我们按照图中?编所画的虚线来调整发射线，调得差不多后就发射攻击对?的建筑，接着建筑倒塌下来会把“?绿猪”给压扁了! 1-2： 根据图中虚线的位置，飞向星球猪下?的建筑?晶。然后，建筑全部垮塌下来，正在嚷嚷地太空猪也随着掉落并消失了。

1-3： 按照图中所?的发射线调得差不多后就发射攻击对?的建筑，接着建筑倒塌下来会把“?绿猪”给压扁了! 1-4：

根据图中所标?的虚线位置，我们点击愤怒的?鸟向左拉后，对准建筑下?的柱?，愤怒的?鸟飞了过去，将柱?撞破后，又碰到了炸药箱，轰的?声，太空猪们被消灭了。 1-5： 根据图中所?的虚线位置，我们将??太空?鸟往左拉对准前?的?晶块，在快要到达?晶块钱，再次点击?晶块(如上图所?)，??太空?鸟会使?冲击波技能将这些零碎的建筑物全部冲击到?猪?上并将它们消灭。 %{p a g e-b r e a k|1-5|p a g e-b r e a k}%

特斯拉Model S电动汽车性能介绍

特斯拉Model S 特斯拉Model S并非小尺寸、动力不足的短程汽车——这是某些人对电动车的预期。作为特斯拉三款电动车中体积最大的车型，根据美国环保署认证，这款快捷、迷人的运动型轿车一次充电能够行驶265英里(426公里)，不过特斯拉声称可以达到300英里。不管哪种情况，这肯定是电动车行业的新高。Model S Performance版本的入门级价格为94,900美元，我测试的版本价格为101,600美元(按照美国联邦税收抵免，可以在此基础上扣减7,500美元)。 在一次开放驾驶上，这款特斯拉汽车硕大的85千瓦时电池的确可以至少行驶426公里。电流来自于车底的电池组，里面有大约7,000颗松下锂电池，重量约为590公斤(1,300磅). 试驾的第二天是前往威斯康辛州，在行驶了320公里后电几乎用光，不过其中包括了在芝加哥的一场交通拥堵中无奈爬行的两个小时。这天的测试充满野心，更多是针对性能而非行驶里程，包括这款特斯拉汽车迅速地用4.4秒时间从0加速到时速97公里(0至60英里每小时)，此外测试达到的最高时速为210公里。 我有没有提到，在0到时速100英里的加速时间方面，这款310千瓦(416马力)的特斯拉汽车将击败威力巨大、使用汽油的413千瓦(554马力)宝马M5？部分原因在于这款特斯拉汽车的同步交流电发动机能够即时提供600牛·米(443英尺磅)的扭矩。像电灯开关一样轻点特斯拉的油门，最大的扭矩已经准备就绪，一分钟内能够实现从0到5,100转。后悬挂、液冷式发动机可以保持1.6万转每分钟，通过一个单速变速箱将动力传导至后轮。 它就像一头冷酷的猛兽，在出奇安静之中让内燃机这个猎物消失于无形——安静到何种程度呢？来自轮胎和风阻的声音比在其他大部分豪华车中感受到的更加明显。安装于车底的电池让特斯拉获得与很多超级车相当的重心，这非常有利于稳定操控。Model S经过弯道的时候也能很好地保持贴地感。 尽管这款特斯拉汽车看起来并不笨重，但其重量达到2,108公斤；随着速度和重力的提升，这些多余的重量表露无遗。加大油门后，沉重的尾部会产生震动。在操控手感的愉悦性方面，特斯拉无法与宝马相提并论，甚至连马自达都赶不上。 美妙的试驾体验在你进入车内之前就开始了，你靠近汽车时，可伸缩的车门把手自动弹出。接着看到的是特斯拉标志性的驾驶室特色内容，一个43厘米(17英寸)电容触摸屏，看起来就像一对相互配合的iPad. 在其用铝合金加强的底盘和车身内，Model S可以容纳5人。一个可爱但是奇怪的按钮可以在车门位置增加脸朝车后的儿童座椅，从而实现最多承载7人。将第二排座椅向下折，可以扩展后座载货空间，可用于家得宝(Home Depot)采购之旅。由于引擎盖下面没有发动机，这些空间可以作为有用的前置行李箱，特斯拉将其称为“前备箱”(“frunk”)，就像保时捷911一样。

三星换货攻略

三星换货攻略 前言：随着大家对东亚换货的熟悉，相信大家都对四星书和二星书的重要性很清楚了，而一星书更适合在东南亚就近根据状态刷食品（番薯）和调味料（蜂蜜、砂糖、罗望子），只有西洋書要在欧洲刷货过来，大家应该也都很熟悉了。而大家都比较忽略三星书的作用，这从三星书的市场价格也能看出来。这里就把三星书的进货和换货路线详细的列举出来供大家参考。 一、欧洲进货篇 1、阿姆刷蜻蜓球or都柏林刷玻璃工艺品，可以拉去云台山、汉阳、浦项、长崎等港 口，这个是大家三星书最常规的用法，这里不多说了。 2、加来刷葛布：这个要求有个法国刷货号。葛布可以拉去杭州，无状态可以5船满比； 也可以拉去堺和江户，这三个城市通常纺织品库存都很低。 3、热那亚双刷古代美术品和大理石像。为什么会提到这个，因为这个是美术品，大家 都知道美术品库存通常都在低位，古美可以在朝鲜和台湾，大理石像可以通吃华南。 4、欧洲进货到此为止，没有列出去贝鲁特刷马刀是因为东地里面太远了，到热那亚相 信已经是一些人的极限了。 二、东南亚进货篇 东南亚可以用三星？是的，没错。这一部分才是本文的重点，熟悉这一部分将对你驰骋南蛮贸易如虎添翼。 有状态时：1、安文炮弹对云台山、杭州、泉州、堺水灾。改版前炮弹是中枢才有，现在不用中枢就可以买到了，投资100万就能看到。经实测在云台山可以4船满比古墨。 2、占碑短剑、马斯喀特弯刀对朝鲜水灾。弯刀基础兑换率是短剑的一倍，如果在印度遇到朝鲜水灾时，可以考虑过去刷。 3、嘉定生丝对各地造船所需，注意去除纤维高库存城市。大部分人造船所需都选择工业制品，而很少选择纤维。经实测生丝可以达到满比兑换。 4、占碑石像对各地经济繁荣。除堺美术品自动高位，其它城市经济繁荣时如果美术品很低，未避免撞车都可以将石像作为一个备选项，基础兑换率在50/20左右。 无状态时：1、泗水金工对朝鲜。基础兑换率为50/30，汉阳工艺品低位时，可以一船满比，其它4船不会低于4：1兑换率。 2、洛布里丝绸布料对杭州。基础兑换率在50/22左右，因为杭州纺织品常年低位，经实测可以一船3：1，其它几船不低于5：1。

三星品牌重塑全攻略(doc22)

三星品牌重塑全攻略(doc22) 【案例提示】 在电子行业进展中，提及在近几年品牌价值提升最快的公司，无疑人们想到的是韩国三星！这家公司成立于1969年，早期业务要紧以生产廉价产品为主，在西方人心目中三星曾经是代表着“低价位，低质量、仿制品”的二三流公司。1990年代，三星施行品牌重塑策略，全力打造三星一流品牌，到现在的三星代表“时尚、高档、技术领先，e化”的全球领导性的品牌公司的蜕变，仅用了36年的时刻！ 时至今日，作为主品牌的三星，其要紧业务范畴包括半导体、数字媒体、通讯网络及数字应用业务，同时在各个领域上三星品牌都有其鲜亮的产品个性，在高端市场上立足，引领时尚潮流。从1990年代的重新塑造三星品牌形象到推出一流产品品牌，从一个经历亚洲金融危机，负债1 70多亿美元，处在破产边缘的公司，到2000年，品牌价值为52亿美元，居世界43位；直至2005年10月27日出版的美国《商业周刊》刊登的品牌价值排行榜上，三星的品牌价值以149亿美元位居“2005年全球100个最有价值品牌”第20位，同时首次超过了它多年的宿敌——索尼（108亿美元，第28位）。三星在电子市场上走出了一条全新的重振品牌之路，并借此跻身全球闻名企业。三星的品牌价值在过去6年增加了186%（如下图所示），制造了增速最快的纪录。 三星是如何摆脱逆境，迅速提升其品牌价值的？它成功的要领在哪里？我们又能从中得到哪些启发呢？针对这些咨询题本文通过对三星的详细研究，首次从品牌价值的角度，揭示了其品牌价值迅速提升的全过程。 攻略一：确定重塑品牌的战略部署 背景：打造强势品牌前的三星状况 在上世纪八十年代末至九十年代初，三星制造的微波炉堆积成山，不得不打折处理，结果降低了自己的品牌声誉，在美国消费者心目中留下二流甚至三流的品牌印象。从而使得三星代表着“低价位，低质量、仿制品”。

特斯拉纯电动车

目录 一、特斯拉简介 (3) 二、特斯拉纯电动车主要功能特点 (3) （一）Model S 主要特点 (3) （二）Model X 主要特点 (9) （三）Model 3 主要特点 (12) 三、特斯拉的电池技术 (13) （一）特斯拉动力电池简介 (13) （二）85kwh电池板的拆解分析 (14) （三）单体电池的能量密度 (20) （四）电量的衰减性能 (22) （五）电池检测实验室：从源头保证锂电池单体一致性 (24) （六）动力电池系统PACK技术 (25) （七）电池管理系统（BMS） (27) 四、特斯拉的充电技术 (35) （一）家用充电桩 (35) （二）超级充电桩 (37) （三）目的地充电桩 (38) （四）计划使用太阳能为超级充电站供电 (38) 五、电机及电控的主要技术 (38) （一）感应电机与永磁电机的对比 (39) （二）Model S采用三相交流感应电机 (40)

（三）双电机可以有效减少高速时的效率降低，并延长续航能力 (41) （四）电机的结构改进提效并易于自动化 (41) （五）逆变器采用分散塑封IGBT，实现低散热要求 (43) 六、车身的主要技术 (46) （一）全铝车身 (46) （二）Model X的双铰链鹰翼门 (47) 七、安全方面的主要技术 (48) （一）车身的安全设计 (49) （二）电池的安全性 (50) （三）信息安全技术 (51) 八、智能化技术 (51) （一）空中升级 (51) （二）远程诊断 (52) （三）自动求助 (52) （四）交互关系 (52)

特斯拉纯电动车的核心技术分析 一、特斯拉简介 特斯拉(Tesla)，是一家美国电动车及能源公司，产销电动车、太阳能板、及储能设备。总部位于美国加利福尼亚州硅谷帕洛阿尔托（Palo Alto）。 特斯拉第一款汽车产品Roadster发布于2008年，为一款两门运动型跑车。2012年，特斯拉发布了其第二款汽车产品——Model S，一款四门纯电动豪华轿跑车；第三款汽车产品为Model X，豪华纯电动SUV ，于2015年9月开始交付。特斯拉的下一款汽车为Model 3，首次公开于2016年3月，并将于2017年末开始交付。 2016年11月17日特斯拉电动车收购美国太阳能发电系统供应商SolarCity，使得特斯拉转型成为全球唯一垂直整合的能源公司，向客户提供包括Powerwall能源墙、太阳能屋顶等端到端的清洁能源产品。2017年2月1日，特斯拉汽车公司（Tesla Motors Inc．）正式改名为特斯拉（Tesla Inc．）。这意味着汽车不再是特斯拉的唯一业务。 二、特斯拉纯电动车主要功能特点 （一）Model S 主要特点 得益于特斯拉独特的纯电动动力总成，Model S 的性能表现十分出色，0-100公里/小时加速最快仅需2.7 秒。通过Autopilot 自动辅助驾驶（选装），Model S 还可以使高速公路驾驶更为安全且轻松，让你更好的享受驾驶乐趣。